Update Article: 57

Original Title

Sanitized Title

Clean Title

Source ID

Article Id01

Article Id02

Corpus ID

Dup

Dup ID

Url

Publication Url

Download Url

Original Abstract

The Court’s extraordinary solicitude for religious expression, manifested across a series of cases involving free exercise, free speech, and establishment clause principles, has been the subject of sustained scholarly attention. Much of that research has focused on evaluating whether the Court is drawing an appropriate balance between the rights of religious believers and government regulatory objectives. In this Article we observe that the Court’s most recent set of moves in this arena, which diverge considerably from decades of speech jurisprudence, will have ramifications that go well beyond the claims of conscience that have so animated the Court’s sympathies. The range of “speakers” protected by this expansive jurisprudence will include information technology companies that generate algorithms and artificial intelligence – speech producers with no conscience at all, much less the kind of sincere religious conviction that the Roberts Court has seen fit to protect against government regulation. As we demonstrate, the free expression principles the Court has developed for religious believers, when added to the Court’s expansive reading of free speech more generally, will make it exceedingly difficult to protect against the significant harms that these speech-producing technologies can cause – including to speakers and readers whom the Court might wish to enable regulators to protect

Clean Abstract

Tags

Original Full Text

University of Denver Digital Commons @ DU Sturm College of Law: Faculty Scholarship University of Denver Sturm College of Law 2-28-2024 From Gods to Google: How Religious Speech Cases May Fortify the Deregulatory First Amendment Rebecca Aviel University of Denver, rebecca.aviel@du.edu Margot E. Kaminski Toni M. Massaro Andrew Keane Woods Follow this and additional works at: https://digitalcommons.du.edu/law_facpub  Part of the Law Commons Recommended Citation Rebecca Aviel, Margot E. Kaminski, Toni M. Massaro, & Andrew Keane Woods, From Gods to Google: How Religious Speech Cases May Fortify the Deregulatory First Amendment, 134 Yale L. J. (forthcoming Feb. 2024). This Article is brought to you for free and open access by the University of Denver Sturm College of Law at Digital Commons @ DU. It has been accepted for inclusion in Sturm College of Law: Faculty Scholarship by an authorized administrator of Digital Commons @ DU. For more information, please contact jennifer.cox@du.edu,dig-commons@du.edu. From Gods to Google: How Religious Speech Cases May Fortify the Deregulatory First Amendment Comments false Publication Statement Copyright held by the author. User is responsible for all copyright compliance. Originally published as Rebecca Aviel, Margot E. Kaminski, Toni M. Massaro, & Andrew Keane Woods, From Gods to Google: How Religious Speech Cases May Fortify the Deregulatory First Amendment, 134 Yale L. J. (forthcoming Feb. 2024). Publication Statement Copyright held by the author. User is responsible for all copyright compliance. Originally published as Rebecca Aviel, Margot E. Kaminski, Toni M. Massaro, & Andrew Keane Woods, From Gods to Google: How Religious Speech Cases May Fortify the Deregulatory First Amendment, 134 Yale L. J. (forthcoming Feb. 2024). This article is available at Digital Commons @ DU: https://digitalcommons.du.edu/law_facpub/824         Arizona Legal Studies Discussion Paper No. 24-07   From Gods to Google    Rebecca Aviel The University of Denver Sturm College of Law Margot E. Kaminski The University of Colorado, Boulder Colorado Law School Toni M. Massaro Andrew Keane Woods The University of Arizona James E. Rogers College of Law   February 2024  THE UNIVERSITY OF ARIZONA James E. Rogers College of Law Electronic copy available at: https://ssrn.com/abstract=4742179        FROM GODS TO GOOGLE                                     FORTHCOMING   134 YALE L. J. (2024)  Rebecca Aviel * Margot E. Kaminski ** Toni M. Massaro *** Andrew Keane Woods ****   Abstract  The Court’s extraordinary solicitude for religious expression, manifested across a series of cases involving free exercise, free speech, and establishment clause principles, has been the subject of sustained scholarly attention. Much of that research has focused on evaluating whether the Court is drawing an appropriate balance between the rights of religious believers and government regulatory objectives. In this Article we observe that the Court’s most recent set of moves in this arena, which diverge considerably from decades of speech jurisprudence, will have ramifications that go well beyond the claims of conscience that have so animated the Court’s sympathies. The range of “speakers” protected by this expansive jurisprudence will include information technology companies that generate algorithms and artificial intelligence – speech producers with no conscience at all, much less the kind of sincere religious conviction that the Roberts Court has seen fit to protect against government regulation. As we demonstrate, the free expression principles the Court has developed for religious believers, when added to the Court’s expansive reading of free speech more generally, will make it exceedingly difficult to protect against the significant harms that these speech-producing technologies can cause – including to speakers and readers whom the Court might wish to enable regulators to protect.    * Professor of Law and Maxine Kurtz Faculty Research Scholar, University of Denver Sturm College of Law.. ** Professor of Law, Colorado Law School. Director, Privacy Initiative at Silicon Flatirons Center; Affiliated Fellow, Information Society Project at Yale Law School. Fulbright-Schuman Grantee 2024. *** Regents Professor and Professor of Law, University of Arizona College of Law. Executive Director, Agnese Nelms Haury Program. **** Milton O. Riepe Professor of Law & Distinguished Legal Scholar, University of Arizona College of Law. Electronic copy available at: https://ssrn.com/abstract=47421792  FROM GODS TO GOOGLE [28-Feb-24     2   CONTENTS  INTRODUCTION ..................................................................................................... 3 I. WHAT TECHNOLOGY LAW LOOKS LIKE, AND WHY ................................... 7 A. Concrete Examples of Recent Technology Laws ................................................ 7 B. Motivating Tropes and Themes in Technology Policy Debates ...........................10 1. The Mediating Role of Technologies ..................................................................... 10 2. The Central Role of Technological Design .......................................................... 11 3. Accessing Technical Expertise ................................................................................ 12 4. Future-Proofing the Law ......................................................................................... 12 5. Concerns with Protecting Innovation and the Broader Economy ................... 13 6. Power Matters ............................................................................................................ 14 II. RELEVANT FIRST AMENDMENT LAW LEADING UP TO 303 CREATIVE: A STORY OF CHURCHES AND PHARMACEUTICAL MARKETERS ....................14 A. Negative Theory and the Oppressed Speaker Paradigm ...................................15 B. The Expansion of First Amendment Coverage and its Connection to “Negative Theory” and the Oppressed Speaker Paradigm .....................................16 C. The Court’s Skepticism of “Content-Based Distinctions” and Fears of Viewpoint and Speaker Persecution .....................................................................19 D. Compelled Speech and Crises of Conscience .....................................................23 III. 303 CREATIVE: FROM FREE EXERCISE TO FREE EXPRESSION .............25 IV. SPEECH DOCTRINE CRAFTED FOR RELIGIOUS BELIEVERS....................30 A. “Pure Speech” and the Erosion of the Speech/Conduct Distinction .................... 31 B. Conflating Customer Speech and Vendor Speech ..................................................... 37 C. Finding that Government Regulation Conscripts the Conscience .......................... 43 D. From Strict Scrutiny to Categorical Invalidity ............................................................ 46 CONCLUSION .......................................................................................................51  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        3 3 INTRODUCTION  As 2023 came to a close, it became increasingly clear that the addictive and manipulative properties of social media have brought about a health crisis for young people, ranging from dangerous weight loss programs such as the “corpse bride” diet to anxiety, depression, and suicidal ideation.1  The Surgeon General issued an advisory detailing “ample indicators” that social media poses “a profound risk of harm to the mental health and well-being of children and adolescents.”2 As captured by Dr. Michael Rich, the doctor who directs the Digital Wellness Lab at Boston Children’s Hospital, “The internet is a giant hypodermic, and the content, including social media like Meta, are the psychoactive drugs.”3  The consensus that policymakers have a role to play in protecting children against these harms has been surprisingly bipartisan,4 and several states have already enacted laws requiring technology companies to take various design precautions and make certain disclosures when it comes to underage users. But because the “psychoactive drugs” are administered through “images, words, and other modes of expression,”5 technology companies are defending themselves with a new set of weapons forged out of the Court’s First Amendment jurisprudence: the idea that anything qualifying as “pure speech” cannot be compelled by the government, no matter how urgent the harms addressed by the challenged law.  While states might be able to demonstrate that the laws are narrowly tailored to advance a compelling interest in protecting children, the Court's most recent case law casts doubt on whether states will even have an opportunity to do so. Perhaps surprisingly, the case most likely to be wielded in near-future lawsuits is not about technology law at all, but about speech by a religious speaker and how it interfaces with public accommodations law. This past summer, in 303 Creative LLC v. Elenis, the Court refused to assess whether the state of Colorado had a compelling interest in enforcing its anti-discrimination law against a wedding website designer who objected to  1 See Ginia Bellafante, If Your Child Is Addicted to TikTok, This May Be the Cure, N.Y. TIMES (Nov. 17, 2023) https://www.nytimes.com/2023/11/17/nyregion/tiktok-social-media-children-addiction.html. For more general discussion of social media and addiction, see Matthew B. Lawrence, Addiction and Liberty, 108 CORNELL L. REV. 259 (2023); Matthew B. Lawrence, Public Health Law's Digital Frontier: Addictive Design, Section 230, and the Freedom of Speech, J. FREE SPEECH L. (forthcoming). 2 U.S. Surgeon General, Advisory on Social Media and Youth Mental Health,  (May 23, 2023) available at: https://www.hhs.gov/about/news/2023/05/23/surgeon-general-issues-new-advisory-about-effects-social-media-use-has-youth-mental-health.html. 3 Matt Richtel, Is Social Media Addictive? Here’s What the Science Says., N.Y. TIMES (Oct. 25, 2023) https://www.nytimes.com/2023/10/25/health/social-media-addiction.html. 4 A Senate bill introduced in the spring — the Protecting Kids on Social Media Act — which would require companies to verify the age of their users was sponsored by unlikely comrades, Chris Murphy, the Democrat from Connecticut, and the Arkansas Republican, Tom Cotton. S.1291, 118th Cong. (2023). See also Tom Cotton, Brian Schatz, Chris Murphy, and Katie Boyd Britt, We Are Senators From Both Parties. It’s Time to Protect Kids on Social Media, WASH. POST (May 11, 2023) https://www.washingtonpost.com/opinions/2023/05/11/social-media-cotton-schatz-murphy-britt/.   5 303 Creative LLC v. Elenis, 600 U.S. 570, 587 (2023). Electronic copy available at: https://ssrn.com/abstract=47421794  FROM GODS TO GOOGLE [28-Feb-24     4 providing services for same-sex couples; the Court instead announced the categorical proposition that the state simply cannot “compel speech that Ms. Smith does not wish to provide.”6  In this Article we explain the profound ramifications of this and other doctrinal shifts for a wide range of regulatory responses to technology-related harms. In the past several years, companies have raised more or less successful First Amendment challenges to attempts to make content moderation more transparent;7 to regulation of the distribution of nonconsensual pornography;8 to facial recognition law;9 to children’s privacy law;10 to broadband privacy law;11 and to law enforcement efforts to decrypt cell phones.12 Whatever one may think of the validity of each of these cases, it is clear that the scope of First Amendment challenges has expanded—from regulation of content, to regulation of all things digital—and that potential First Amendment pitfalls are abundant to the point of swallowing effective tech regulation whole. The doctrinal writing is on the wall that companies will raise similar challenges to attempts to regulate artificial intelligence (AI).13 Our point is not to defend any of these regulations here; rather, our point is that in each of these domains, the specter of the First Amendment looms as a nearly insurmountable barrier to government regulation. And this is not merely a story of a generally expanding and deregulatory First Amendment. It is a more specific story of how the current Supreme Court’s ongoing solicitude for what it sees as minoritized religious speakers is creating a cudgel for technology companies to challenge what are often bipartisan technology regulations. We aim to help the reader understand both the developing narrative and doctrine of the First Amendment, and why technology regulations are particularly susceptible to free speech challenge, especially in the wake  6 Id. at 588. 7 This month, the Court is hearing two cases that challenge state content moderation laws on First Amendment grounds: NetChoice, LLC v. Paxton, No. 22-555, 2023 WL 6319650 (U.S. Sept. 29, 2023) and Moody v. NetChoice, LLC, No. 22-277, 2023 WL 6319654 (U.S. Sept. 29, 2023). 8 State v. VanBuren, 214 A.3d 791, (Vt. 2019) (overruling lower court decision dismissing charges under state revenge porn statute on the theory that it violated the First Amendment); People v. Austin, No. 123910, 2019 WL 1870855 (Ill. 2019) (same). 9 ACLU v. Clearview AI, Inc., No. 20 CH 4353, 2021 WL 4164452 (Ill. Cir. Ct. Aug. 27, 2021) (applying intermediate scrutiny to find that Illinois’s facial recognition statute does not violate the First Amendment).  10 NetChoice v. Bonta, 2023 WL 6175551 (N.D. Cal. Sept. 18, 2023) (granting trade association’s motion for preliminary injunction challenging California Age Appropriate Design Code Act on First Amendment grounds).   11 ACA Connects v. Frey, 471 F.Supp.3d (318) (2020) (denying a motion to dismiss internet service providers’ First Amendment challenge to Maine’s online privacy statute). 12 In re Search of an Apple iPhone Seized During the Execution of a Search Warrant on a Black Lexus IS300, No. ED 15-0451M, 2016 WL 618401, at *1 (C.D. Cal. Feb. 16, 2016). 13 See Toni M. Massaro & Helen Norton, Siri-ously? Free Speech Rights and Artificial Intelligence, 110 NW. U. L. REV. 1169 (2016) (exploring how First Amendment theory and doctrine both leave room for protecting non-human speech); Toni M. Massaro, Helen Norton, and Margot E. Kaminski, SIRI-OUSLY 2.0: What Artificial Intelligence Reveals About the First Amendment, 101 MINN. L. REV. 2481 (2017) (explaining that the First Amendment may protect strong AI speech, but that this need not deprive free speech of a human-centric focus). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        5 5of 303 Creative. That is, we aim to draw the through line from gods to Google (and TikTok and OpenAI and…). Regulators and scholars have long understood that the First Amendment is the iceberg around which all technology regulations must navigate.14 Scholars in privacy, cybersecurity, robotics, antitrust, and more increasingly see the First Amendment as a significant barrier to effective regulation, with widespread anxiety about the so-called Lochnerization of the First Amendment.15 If a corporation’s actions are speech, if personal data are speech, and if machine outputs are speech, then nearly every technology regulation issue is also a free speech issue.   Technology law scholars and policymakers may be highly attuned to the consequences of more obviously applicable First Amendment cases, like Sorrell v. IMS Health Inc., which addressed a data privacy law of sorts,16 or the upcoming NetChoice v. Paxton, addressing online content moderation.17 There are hundreds—perhaps thousands—of articles addressing Sorrell, in which the Supreme Court in dicta suggested that “data is speech.”18 A whopping seventy-eight organizations submitted amicus briefs in Gonzalez v. Google, a 2023 case implicating content moderation policy through Section 230 of the Communications Decency Act.19 But few technology law scholars are likely to be paying close attention to the Court’s religious speaker doctrine, including its most recent case, 303 Creative. That is a mistake. The Court’s religious speaker cases generally favor speakers over regulators, and they have fundamentally changed modern First Amendment analysis in ways particularly consequential for technology regulation. In particular, the caselaw now has set forth a categorical rule prohibiting anything that can be plausibly characterized as compelled “pure speech” – no matter how urgent the harms at issue. This is a recent and frightening development.  How did we get here? When did the urgency of a regulator’s objectives cease to be a relevant component of the free speech assessment?  In this Article, we explain why the religious speaker cases logically offer Meta, Google, and other technology  14 See, e.g., Robert Post, Encryption Source Code and the First Amendment, 15 BERKELEY TECH. L.J. 713 (2000); Andrew Tutt, Software Speech, 65 STAN. L. REV. ONLINE 73 (November 15, 2012); Alan Z. Rozenshtein, Silicon Valley’s Speech: Technology Giants and the Deregulatory First Amendment, 2021 J. FREE SPEECH L. 337, 340 (2021) (“[A]n important commonality across these regulatory areas is the potential for the First Amendment to act as a potent tool against government regulation.”). 15 See, e.g., Frederick Schauer, The Boundaries of the First Amendment: A Preliminary Exploration of Constitutional Salience, 117 HARV. L. REV. 1765, 1787–93 (2004); Amanda Shanor, The New Lochner, 2016 WIS. L. REV. 133 (2016); Jeremy K. Kessler, The Early Years of First Amendment Lochnerism, 116 COLUM. L. REV. 1915 (2016); Robert Post and Amanda Shanor, Adam Smith’s First Amendment, 128 HARV L REV F. 165, 182 (2015). Cf. Genevieve Lakier, The First Amendment’s Real Lochner Problem, 87 U. CHI. L. REV. 1241 (2020) (arguing that “contemporary free speech law is not Lochner-like in failing to defer to the legislature’s economic policy decisions. Instead, it repeats the errors of the Lochner Court by relying upon an almost wholly negative notion of freedom of speech and by assuming that the only relevant constitutional interest at stake in free speech cases is the autonomy interest of the speaker”). 16 564 U.S. 552 (2011). 17 2023 WL 6319650. 18 A cursory search on Westlaw shows over 1200 law review articles citing the case. 19 See Gonzalez v. Google, LLC, No. 21-1333 (collecting Amicus briefs) https://www.supremecourt.gov/docket/docketfiles/html/public/21-1333.html. Electronic copy available at: https://ssrn.com/abstract=47421796  FROM GODS TO GOOGLE [28-Feb-24     6 companies enhanced power to resist regulation. In its skepticism about the scope of modern anti-discrimination law and its sympathy for religious claimants who assert that these laws cannot be applied to them, the Court has expanded rights that cannot be contained to religious dissidents. These developments are regrettable on their own terms, as other scholars have noted in what is sure to be a growing chorus: it is difficult to see how public accommodations laws can survive “the vast and careless overreach of the concept of ‘pure speech’” that Justice Gorsuch developed in 303 Creative.20 In this Article we urge an even broader assessment of the Court’s eagerness to protect religious speakers: it is not just public accommodations laws that are imperiled by the uncontainable logic of these cases. It is the wide array of regulatory responses that sensible policymakers might choose to deploy against the acute and rapidly proliferating risks we face from technology.  To understand this impending collision, one must recognize both relevant features of technology regulation, and the recurring narrative driving the Court’s religious speaker jurisprudence and related cases. In its recent First Amendment cases, driven largely but not solely by Justice Thomas, the Court has clutched to a recurring story of persecuted religious minorities standing fast in the face of domineering and ideologically majoritarian regulators. It invokes as its touchstone the Jehovah’s Witness schoolchildren forced to affirm the Pledge of Allegiance by a public school.21 That 1943 religious speaker case, Barnette, is the lens through which the Court, and lower courts, will be trying to parse complex regulation, including of AI systems. In other words, the Court is about to try to read technology law through a doctrine calibrated to protect persecuted religious individuals—to construct Google as a modern-day Jehovah’s Witness forced to recite the Pledge of Allegiance while saluting the flag. The Article begins by explaining, in broad strokes, what current technology regulation often looks like and why, including platform regulation, data privacy, and laws governing AI. This summary sets the stage for our argument that First Amendment religious speaker doctrine is particularly dangerous for contemporary technology law. We offer an overview of First Amendment cases leading up to 303 Creative in which we can see the development of the oppressed speaker paradigm.  We then turn to 303 Creative in particular. We identify five moves in the opinion that together threaten the viability of much technology regulation: 1) choosing free speech over free exercise to decide the case; 2) eliding the speech-conduct distinction in analyzing a public accommodations law; 3) treating the speech at issue as the commercial vendor’s own in ways that resemble a compulsory flag salute; 4) treating any compliance burden as undue coercion, regardless of the expressive alternatives available; and 5) applying a rule of categorical invalidity rather than engaging strict scrutiny.  For each, we explain the ramifications for sensible technology policy.  We conclude with a sober reminder as we head into yet another highly consequential election: if we cannot pursue nimble, sophisticated, and effective responses to the ways  20 Robert Post, Public Accommodations and the First Amendment: 303 Creative and “Pure Speech”, draft available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4571189. See also David S. Schwartz, Making Sense of 303 Creative: a Free Speech Solution in Search of a Problem, (arguing that “the 303 Creative Court blunders into this problem [of whether public accommodations laws can command compliance that is expressive of ideas with which a person disagrees], and addresses it partially, myopically, and badly). 21 West Va. State Bd. of Educ. v. Barnette, 319 U.S. 624 (1943).  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        7 7that new technology can be deployed to undermine the electoral process, our democracy will struggle to be worthy of the name.    I. WHAT TECHNOLOGY LAW LOOKS LIKE, AND WHY  To understand the throughline from gods to Google it is essential to understand the shape and purpose of recent technology law. “Technology law” can mean any number of things, but we write primarily of U.S. law governing digital information flows. This includes aspects of communications law, data privacy law, and the quickly developing law of AI.22 These laws govern technologies built of information, often for information-related purposes. These technologies may gather, host, distribute, process, channel, mediate, and sell information; as we explain, they are not merely analogous to the newspapers and pamphlets of yesteryear.  Technology law reflects a number of recurring policy moves.  They typically contain one or more of the following: (1) transparency requirements, including disclosures to affected individuals; (2) accountability to experts, such as auditing; and (3) measures aimed at influencing technological design. Recent technology laws in the United States often are not omnibus in nature, but are (4) aimed at a particular subset of actors or sector of industry most likely to cause harms. They are also often (5) written using broad and technology-neutral terms, thus delegating substantial interpretative discretion to either a regulatory agency or the private sector, in the name of future-proofing the law or deferring to private sector expertise. This Part outlines these essential technology law components. The aim of this Part is descriptive. We are not arguing that these are good technology policy choices, rather that these are the dominant policy instruments of the day. As we will show, these policy choices are particularly vulnerable to First Amendment challenges.  A.  Concrete Examples of Recent Technology Laws  We offer first a few concrete examples to illuminate technology law’s broad themes. Take data privacy, which is related to, but distinct from, the kind of privacy many people first think of: a right to be left alone, to be free from recording or photographs of solitary or confidential moments. In fact, data privacy governance arises from the now-pervasive practice of sharing private information with an intermediary that may not have your best interests in mind.  Data privacy in the United States has always centered on transparency requirements. “Notice-and-choice” regulation, widely criticized for its inadequacies, has been the dominant approach to data privacy in the United States for decades.23 Notice-and-choice aims to ensure that privacy disclosures provide users  22 Intellectual property law could be on this list, too, but as it has its own structure and logic and sui generis First Amendment interface. 23 See e.g. Daniel J. Solove, The Limitations of Privacy Rights, 98 NOTRE DAME L. REV. 975, 998 (2023) (“In the United States especially, privacy law relies far too heavily on the notice-and-choice approach, which involves providing people with notice and then relying on them to make decisions about their privacy.”); Neil Richards & Woodrow Hartzog, Taking Trust Seriously in Privacy Law, 19 STAN. TECH. L. REV. 431, 444 (2016) (“When the FTC first started to regulate privacy in the late 1990s, it adopted a basic notice and choice regime for businesses that was congruous with many of the FIPs.”); Woodrow Hartzog & Neil Richards, Privacy’s Constitutional Electronic copy available at: https://ssrn.com/abstract=47421798  FROM GODS TO GOOGLE [28-Feb-24     8 sufficient notice to effectuate individual choices about what data technologies to use.  Starting in 2018, a wave of states—including California, Colorado, Virginia, Utah, and more—have responded to the inadequacies of notice-and-choice approaches to data privacy by enacting laws that regulate the collection, processing, and use of personal data.24 These laws are typically directed at a small subset of businesses most likely to have large impacts.25 The dual cores of these laws are: (1) a set of individual rights that center on transparency and individual choice, and increasingly (2) an accompanying light-touch regulatory regime that is based on “risk management.”26  The original California data privacy law, the California Consumer Privacy Act (CCPA), was fundamentally a transparency law.27 Like other laws, the CCPA emphasized the individual’s “right to know,” imposing both public-facing transparency and individualized transparency of varying kinds.28 Individuals are to be alerted, for example, if a company is collecting their data, which of their personal data a company has, how companies are going to use that data, including whether they intend to sell it, and for what reasons. Increasingly, regulators have also mandated transparency about these transparency disclosures, requiring for example that they be posted in clearly understandable terms.29 Transparency rights are intended to be usable—to effectuate other specific individual rights, such as a right to opt out of the sale of one’s data, or a right to opt out of profiling, or a right to correct mistakes in one’s data. These individual rights, based on a set of principles that date back to the 1970s (the “FIPPs”),30 aim to correct power imbalances and institute procedural fairness. They form the core of data protection laws around the world.  The regulatory component of these newer data privacy laws typically uses an  Moment, 61 B.C. L. Rev. 1681, 1704 (2020) (““notice” often means little more than burying data practices in the fine print of a dense privacy policy, while “choice” means choosing to use a service with its non-negotiable data practices as a take-it-or-leave-it option.”). 24 See, e.g., The California Consumer Privacy Act of 2018, Cal. Civ. Code § 1798.100-199.100 (regulating the collection and processing of consumer data); Colorado Privacy Act, 4. Colo. Code Regs. § 904-3-7.09 (regulating the collection and use of personal data).  25 California Consumer Privacy Act of 2018 (https://oag.ca.gov/privacy/ccpa) businesses definition: defined by size (really, annual revenue), amount of personal data processed (typically, of over 100k people), and/or the extent to which their business models rely on the selling or sharing of personal data. 26 Margot E. Kaminski, Binary Governance: Lessons from the GDPR’s Approach to Algorithmic Accountability, 92 S. CAL. L. REV. 1529, 1552–53 (2019); Ari Waldman, The New Privacy Law, 55 U.C. DAVIS. L. REV. ONLINE 19 (2021); Anupam Chander, Margot E. Kaminski & William McGeveran, Catalyzing Privacy Law, 105 MINN. L. REV. 1733 (2021). See also Jane R. Bambauer, How to Get the Property Out of Privacy Law, draft available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4629539 (calling for a move to more risk regulation and fewer individual rights). 27 Chander, Kaminski, McGeveran, supra note 26 (calling the CCPA a uniquely American data privacy law). 28 Id. See also Solove, Limitations of Privacy Rights, supra note 23 (discussing the right to know). 29 See, e.g., California Age-Appropriate Design Code Act, Cal. Civ. Code § 1798.99.31(a)(7) (requiring that disclosures be clear and understandable through the use of age-appropriate language in disclosures). 30 See Woodrow Hartzog, The Inadequate, Invaluable Fair Information Practices, 76 MD. L. REV. 952, 953 (2017) (“The FIPs model of privacy regulation has been adopted by nearly every country in the world that has decided to take data protection seriously.”). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        9 9instrument known as an “Impact Assessment” (also called a Data Protection Impact Assessment (DPIA) or Data Privacy Impact Assessment (DPIA)). This instrument requires a company to assess the risks of a particular surveillance or data processing technology before and sometimes during its use, and often to mitigate those risks ex ante. Again, data privacy laws around the world have used versions of this light-touch risk regulation, tasking companies with policing themselves while also trying to design various forms of accountability over that delegation to prevent it from turning into capture. At the same time, these laws use broad terms in defining possible rights—terms like “privacy”—effectively delegating decisions about their interpretation and application to the regulated entities, both to leave room for unforeseen challenges and to incorporate private sector expertise. New data privacy laws also reflect what is known as “Privacy by Design”: an attempt to bake privacy values into technological and organizational design early in the development of a technology.31 This acknowledges the reality that technological interfaces are often deliberately designed to be viscerally manipulative and even addictive. The recent California children’s privacy law, modeled after United Kingdom regulation, is so directly aimed at design that it is called the “California Age-Appropriate Design Code Act.”32 Several recent U.S. privacy laws more specifically target “dark patterns,” attempts to manipulate users against their own interests often through addictive interface design.33  Use of these regulatory tools in recent technology law—transparency, risk assessments, design parameters, targeting of specific entities to prevent impacts on broader innovation, the use of broad terms to try to future-proof the law—is not limited to data privacy laws. Delegated risk management, coupled with individual process rights, has also been a core component of proposals to regulate AI, and of recent attempts to impose “fairness” on platform content moderation in the United States. AI risk regulation tasks companies and/or government entities that use AI or similarly opaque algorithms with evaluating ex ante what the risks will be for an array of concerns, from discrimination and fundamental rights to safety.34 The so-called “Algorithmic Impact Assessment” has been central to many proposed AI laws and guidance, from Singapore to the EU AI Act, to the U.S. self-regulatory regime. AI risk regulation also typically tasks companies with solving identified problems themselves, subject to light-touch oversight (if that). Regulators have been turning to external expert audits to  31 This approach was adopted by the FTC as early as 2012. See Remarks of Commissioner Edith Ramirez, Privacy By Design and the New Privacy Framework of the U.S. Federal Trade Commission, June 13, 2012, available at: https://www.ftc.gov/sites/default/files/documents/public_statements/privacy-design-and-new-privacy-framework-u.s.federal-trade-commission/120613privacydesign.pdf. 32 Cal. Civ. Code § 1798.99.29. See also NetChoice v. Bonta, 23-2969 (9th Cir. 2023) (brief for appellant Bonta) (noting that the CAADCA is “modeled after the United Kingdom’s Age Appropriate Design Code, commonly referred to as the ‘Children’s Code,’ which requires that all websites likely to be accessed by children provide privacy protections by default”). 33 See, e.g., Colorado Privacy Act, 4. Colo. Code Regs. § 904-3-7.09 (“Consent obtained in violation of this part 4 CCR 904-3, Rule 7.09 may be considered a Dark Pattern, as defined in C.R.S. § 6-1-1303(9).”). 34 See Margot Kaminski on Regulating AI Risks, The Lawfare Podcast (April 20, 2023), https://www.lawfaremedia.org/article/the-lawfare-podcast-margot-kaminski-on-regulating-ai-risks. Electronic copy available at: https://ssrn.com/abstract=474217910  FROM GODS TO GOOGLE [28-Feb-24     10 establish further accountability. Some have looked to individualized transparency rights (a “right to explanation” of an AI decision) and accompanying individual substantive rights (a “right to contest” AI decisions) that are proposed as partial solutions to algorithmic opacity and shifts in power dynamics away from the affected individual.35 Even recent content moderation laws deploy similar tools.36 People tracking content moderation policy have experienced abrupt whiplash as the Republican Party shifted from a hands-off approach towards tech behemoths to importing aspects of EU-style regulation.37 In the name of fairness and consistency, new-wave content moderation laws require transparency about moderation policies (analogous to disclosure of a privacy policy), and sometimes individualized explanations of decisions.  First Amendment questions about these laws already abound: whether states can require tech companies to provide individualized explanations for their decisions to remove, edit, or arrange content on their platforms is currently before the Supreme Court in NetChoice v. Paxton and Moody v. NetChoice.38 And in California, a judge struck down nearly every element of the new children’s privacy law, including impact assessments and requirements that disclosures be comprehensible to the children at whom they are aimed, on First Amendment grounds.39   B.  Motivating Tropes and Themes in Technology Policy Debates  This slate of approaches arose in response to particular recurring concerns that shape technology policy conversations. Here we outline core themes in technology policy leading to the use of these tools—especially the heavy reliance on transparency and disclosures in governing information technology. We focus at a high level on six central themes: (1) the mediating role of the platform and other technologies; (2) the central role of technological design; (3) the puzzle of how to incorporate non-legal expertise into the law; (4) the problem of future-proofing the law; (5) a concern with avoiding harms to innovation; and (6) an increasing focus on distribution of power.  1. The Mediating Role of Technologies  One central theme of contemporary technology law is the mediating role of the platform and other technologies. When technology law talks of “platforms” it typically means online services that host user content, such as social networks or search engines; but many other technologies can be construed as playing a mediating role—including one’s cell phone, facial recognition, and most recently Large Language Models such as Chat GPT. By “mediating” we mean that the platform is in fact a platform: it sits between  35 For an overview, see Margot Kaminski and Jennifer Urban, The Right to Contest AI, 121 COLUM. L. REV. 1957, 1975-1984 (2021). 36 For example, the Texas law at issue in NetChoice v. Paxton includes a stringent disclosure requirement aimed at increasing platform transparency. See NetChoice, LLC v. Paxton, supra note 7.   37 The effect of European influence here cannot be overstated: these are the tools common to a number of recent EU regulations, from the General Data Protection Regulation (GDPR) to the Digital Services Act (DSA).  38 Case Nos. 22-277, 22-393, 22-555.  39 NetChoice v. Bonta, 2023 WL 6175551 (N.D. Cal. Sept. 18, 2023). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        11 11citizens and other individuals, citizens and the market, and citizens and their governments (or governments and their citizens). By playing this mediating role, the platform does more than just speak itself. It constitutes individuals and social interactions through mediation and even deliberate manipulation.40  Discussion of the platform as mediator also emphasizes that platforms and other technologies are not neutral. The companies that build them make choices, often self-serving ones. Increasingly, technology regulation has moved from asking the “collateral censorship” question—that is, “how do we avoid over-regulating platforms such that they won’t remove too much user content?”41—to asking the “manipulation” question—that is, “how do we think about regulating platforms so that they won’t manipulate their users towards self-interested ends?”42 Both of these are questions about the alignment and misalignment of platform interests and user interests, and what to do when they diverge.  2. The Central Role of Technological Design  Another longstanding theme in technology policy involves the import of technological design.43 Design governs people: it constrains them, channels them, enables them. Code encodes values. Thus, regulators can and do intervene in the design of technologies all the time. They require car manufacturers to put in seatbelts, for safety. They nudge websites to capture and share user information with the government. They push AI companies to try to minimize risks of disinformation and discrimination. Thorny questions arise about the legitimacy of government intervention in technological design.44 But government does so regularly, both because sometimes it is necessary to protect individuals, and because sometimes it is the most effective way to get what the government wants. Design constrains behavior, so regulating through design can effectively produce government-favored outcomes while potentially hiding government involvement – thus absolving it from immediate responsibility and feedback.45  40 Julie E. Cohen, What Privacy is For, 126 HARV. L. REV. 1904 (2013) (arguing that privacy protects a dynamic intersubjective ”self” as well as political and intellectual culture). 41 See, e.g., Felix T. Wu, Collateral Censorship and the Limits of Intermediary Immunity, 87 NOTRE DAME L. REV. 293 (2013). 42 See, e.g., Ryan Calo, Digital Market Manipulation, 82 GEO. WASH. L. REV. 995 (2014) (adapting and updating behavioral psychology concerns to the digital consumer marketplace). See also Rory Van Loo, Digital Market Perfection, 117 MICH. L. REV. 815 (2018) (discussing manipulation impacts of AI). 43 See, e.g., LAWRENCE LESSIG, CODE AND OTHER LAWS OF CYBERSPACE 30-42 (1999) (describing the way that design choices, like architectural choices, constrain behavior in law-like ways); Joel Reidenberg, Lex Informatica: The Formulation of Information Policy Rules Through Technology, 76 TEX. L. REV. 553, 554 (1997) (“law and government regulation are not the only source of rule-making. Technological capabilities and system design choices impose rules [too]). 44 See, e.g., Deirdre K. Mulligan & Kenneth A. Bamberger, Procurement As Policy: Administrative Process for Machine Learning, 34 BERKELEY TECH. L.J. 781 (2019) (describing the tension between the regulator’s “procurement mindset” and technological design). 45 See, e.g., AARON PERZANOWSKI, THE RIGHT TO REPAIR 110-124 (2021) (describing how intellectual property regimes like the Digital Millenium Copyright Act can give private actors Electronic copy available at: https://ssrn.com/abstract=474217912  FROM GODS TO GOOGLE [28-Feb-24     12 Attempts at regulating design typically are macro, not micro, in nature. That is because design changes quickly, and expertise is often housed in the private sector. The FTC might issue general guidelines on how to design effective notice in mobile apps.46 But legislators on the whole are unlikely to provide more detailed marching orders; instead they offer general principles and give the private sector room to instantiate practical applications of the principles.  3. Accessing Technical Expertise  This brings us to the issue of technical expertise. While pacing is rarely a serious issue–the law deals with technological and other social change all the time–getting governmental access to technological and industry expertise truly is a conundrum. Proposed solutions include the establishment of new expert agencies, funding tech-fellows in Congress or other legislative bodies, and enhancing technological expertise within existing agencies. Policy conversations also involve institutional choices, exploring whether courts, legislatures, or aging and possibly under-resourced agencies are best situated to make big policy decisions and to gain access to relevant expertise. In these discussions we see emphasis on a particular kind of transparency, aimed at moving expertise from within industry to regulators and the public.  Examples include regulatory monitoring47 and required disclosures and even explanations to regulators, individuals, the public, or third-party auditors. Policymakers and scholars call for ways to harness expertise within industry, eschewing top-down command-and-control governance for less traditional forms of governance termed “collaborative governance” or “new governance.”48 This much-critiqued but often-deployed approach to technology law involves partial delegations to private industry, ideally supervised by a regulator with a significant enforcement power but an opportunity for forbearance, and subject to third-party and public oversight to prevent capture.49 Again, transparency is central to these endeavors, as an aspect of accountability over delegation.  4. Future-Proofing the Law  Sometimes the problem technology regulators want to solve is to future-proof the law as best they can against inevitable change, preventing newly emerging companies from trying to claim they are not covered by regulation.50  (One of us has railed against  enormous power to shape their products in ways that do not in fact comport with the goals behind the intellectual property regime). 46 See Federal Trade Commission, Mobile Apps, video available at: https://www.ftc.gov/media/71318. 47 Van Loo, supra note 42. 48 Kaminski, Binary Governance, supra note 26; INNOVATIVE GOVERNANCE MODELS FOR EMERGING TECHNOLOGIES (Gary E. Marchant, Kenneth W. Abbott, & Braden Allenby, eds. 2014) (collecting essays calling for new governance mechanisms to address a range of novel technology issues).  49 See Kaminski, Binary Governance, supra note 26 at 1559-1561 (describing collaborative governance models).  50 See, e.g., UC Davis Law Review Symposium: Future-Proofing Law, 51 U.C. DAVIS L. REV. 1 (2017) (collecting essays describing the challenge of, and proposing some solutions to, the fact Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        13 13the extreme version of this so-called “pacing problem”—the idea that the law cannot keep up with technological development–because there are plenty of tools in the regulatory toolkit for keeping pace with changes of all sorts, technological or otherwise.51) Future-proofing typically involves a variety of regulatory tools. Technology lawmakers often attempt to future-proof a law by incorporating technologically-neutral language, focusing on specific harms or actors or industries or use cases, rather than naming specific and quickly dated technologies.52 They attempt to future-proof by empowering expert agencies, such as the FTC, with broad mandates.53 They future-proof by toggling between rules and standards, perhaps crafting a set of standards and then letting expert agencies or courts develop more detailed rules as new applications arise.54 And they future-proof by instituting (once again!) transparency requirements, in the hope that a news cycle about particularly egregious new harms will lead to further regulation.55  5. Concerns with Protecting Innovation and the Broader Economy  U.S. technology policy conversations often reflect concerns about protecting innovation—whatever that means.56 Policy discussions typically reflect concern about overburdening smaller or newer-stage companies. This again affects institutional choice and design. That is, if you think “clear rules of the road” are better, especially from a competition standpoint, you'd want ex ante regulation rather than ex post. You’d want expert agencies providing clarity for the “little guys” instead of courts, uncertainty, and litigation costs. You would also want narrow tailoring of regulation to actual harms. For better or worse, U.S. tech policy conversations are often anti-precautionary principle. In the United States, unlike in Europe, regulation arises once harms are proven, rather than in anticipation of them. Technology policymakers in the U.S. thus  that laws and technologies do not always arise at the same time). 51 Margot Kaminski, Technological “Disruption” of the Law’s Imagined Scene: Some Lessons from Lex Informatica, 36 BERKELEY TECH. L.J. 883, 890-95 (2021) (describing how the pacing problem is often less of a problem than it seems).  52 See Michael Birnhack, Reverse Engineering Informational Privacy Law, 15 YALE J.L. & TECH. 24 (2012) (outlining methods for designing technology law that is technology neutral and therefore more durable as technologies change). 53 Daniel J. Solove and Woodrow Hartzog, The FTC and the New Common Law of Privacy, 114 COLUM. L. REV. 583 (2014) (describing how the FTC developed a sweeping privacy jurisprudence that can be understood as a body of common law).  54 See, e.g., Mark A. Lemley & Bryan Casey, Remedies for Robots, 86 U. CHI. L. REV. 1311, 1382 (2019) (describing how standards and rules offer different tradeoffs for regulators crafting sensible technology policy). 55 See e.g., A. Michael Froomkin, Regulating Mass Surveillance as Privacy Pollution: Learning from Environmental Impact Statements, 2015 U. Ill. L. REV. 1713 (2015) (making the case for privacy impact notices that could, in theory, ignite a positive regulatory dynamic).  56 For example, the White House’s most recent Executive Order on AI emphasizes “promoting innovation” as among its eight guiding principles. Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, Oct. 30, 2023, available at: https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/.  Electronic copy available at: https://ssrn.com/abstract=474217914  FROM GODS TO GOOGLE [28-Feb-24     14 tend to focus on particular industries, or a narrower class of actors, rather than go omnibus and broad like the General Data Protection Regulation (GDPR) or EU AI Act. In the U.S., technology law very often involves transparency as a first-step regulatory move aimed at airing harms and problems towards possible future regulation.57  6. Power Matters   Much of contemporary technology law is aimed at correcting power imbalances: between platforms and their users, data processors and impacted individuals, hiring companies and employees affected by AI screening tools. Thus much of technology law looks to moves like adding disclosure and process, aimed at giving individuals purchase over vast areas that affect their lives. This all occurs against the backdrop of increasing suspicion that there aren’t many little companies left, with incumbent technology companies exploiting existing positions of power not just against consumers but against competitors in the marketplace. The foregoing themes do not dictate all regulatory outcomes, and again, we do not claim they are all good regulatory moves. But they give some idea of how a more militant application of the First Amendment will remove significant policy tools from the toolkit, greatly limiting lawmakers’ policy choices and squelching attempts to thoughtfully experiment with governance and institutional design. Specifically, many of the most common forms of tech regulation plainly involve information products that increasingly could be characterized as First Amendment covered speech, and often compel disclosure by private actors.  In the following part, we outline the many ways in which the Court has greased the path for these private actors to object to tech regulation on First Amendment grounds, often by relying on cases that involve religious speakers, or invoke related paradigms.  II. RELEVANT FIRST AMENDMENT LAW BEFORE 303 CREATIVE: A STORY OF CHURCHES AND PHARMACEUTICAL MARKETERS  Several doctrinal developments in First Amendment law portend devastating consequences for the technology laws discussed above, including the Court’s steady expansion of First Amendment coverage; its tendency to invoke some form of heightened scrutiny of ordinary economic regulation when it perceives even a sniff of government content- or speaker-favoritism; and its growing sensitivity to any form of government “compelled speech.” We argue that all of these developments can and should be read through the lens of one recurring motivating logic: a desire to protect minority speakers—often religious—from dogmatic, controlling, and even persecuting legislatures. Echoes of this narrative pervade recent First Amendment caselaw. The deregulatory turn of First Amendment doctrine in recent years is well documented.58 But to our eyes, there is a missing link between that turn and the Court’s solicitude for what it believes are oppressed religious speakers.   57 See Andrew D. Selbst, An Institutional View of Algorithmic Impact Assessments, 35 HARV. J.L. & TECH. 117, 158 (2021) (describing transparency’s use as an accountability mechanism). 58 See sources discussed in note 15, supra.  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        15 15We begin this Part by describing the Court’s turn to a negative theory of speech protection, and how that theory has been tethered to a recurring narrative of an oppressed and often religious speaker who defies majority views. This narrative has fueled a basic First Amendment justification—that free speech law operates as an essential bulwark to protect minoritized speakers and viewpoints from overreaching by majoritarian (and now typically, liberal) legislatures. This Court, with its supermajority that unites both social conservatives and free-market conservatives, uses that narrative to justify extending speech protections to speakers—like pharmaceutical marketers—who might seem far from the motivating narrative.  A.  Negative Theory and the Oppressed Speaker Paradigm  A negative theory of the First Amendment is grounded on a particular view of what free speech principles accomplish for society.  It focuses not on the affirmative benefits we might enjoy from robust and unhindered speech but rather the negative consequences that arise from government regulation.  Negative theory is powerful because, as Helen Norton observes, “the government gives us plenty of reason to distrust it.” 59 Because it seeks primarily to curb government overreach rather than to promote democratic participation or ensure a marketplace of ideas, a negative theory approach begins from a place of skepticism regarding government interests and is thus inherently hostile to regulation.60 Without limiting principles, it eats government regulation whole.61 The free speech jurisprudence of the Roberts Court reflects the animating concerns of negative theory.  It is preoccupied with preventing government from repressing and conscripting minoritized speakers in the name of majority interests. This “Don’t Tread on Me” instinct dominates regardless of the strength of the government’s regulatory interests, and regardless of how general-purpose or facially neutral a law might appear. Our observation here is that negative theory now comes with a particular paradigmatic narrative—one that increasingly links it to religious speakers and religious speech. That is the narrative of the oppressed minority speaker, standing fast in the face of an overreaching and dogmatic majority government. For this Court, particularly Justice Thomas, that paradigmatic speaker is religious and socially conservative: a  59 See Helen Norton, Distrust, Negative First Amendment Theory, and the Regulation of Lies, 22-07 KNIGHT FIRST AMEND. INST. (Oct. 19, 2022), https://knightcolumbia.org/content/distrust-negative-first-amendment-theory-and-the-regulation-of-lies)(“First Amendment law exemplifies this tradition of distrust… the First Amendment tradition also relies on what many call a negative theory of the Free Speech Clause. Under this approach, the Constitution protects speech not so much because it is so valuable, but instead because the government is so dangerous in its capacity to abuse its regulatory power.”). See also Massaro, Norton, and Kaminski, SIRI-OUSLY 2.0: What Artificial Intelligence Reveals About the First Amendment, supra note 13(discussing how the implications of this expansion of theoretical support may lead logically, albeit not inevitably, to protection of much AI speech); Lakier, supra note 15 (arguing that the Court’s “Lochner” turn is based on the embrace of negative theory and emphasis on speaker autonomy). 60 Id. 61 Norton, supra note 59 (proposing limiting principles for negative theory  and observing that“negative theory serves as a guardrail on government, but negative theory warrants guardrails of its own to prevent the paralysis that accompanies unbounded distrust.”). Electronic copy available at: https://ssrn.com/abstract=474217916  FROM GODS TO GOOGLE [28-Feb-24     16 church that can’t post signs for its services,62 an anti-abortion protestor who is required to keep a physical distance from an abortion clinic,63 a Catholic Church that can’t post signs about confession,64 a faith-based “crisis pregnancy center” that is compelled against its conscience to inform women coming in for healthcare that it provides no abortions.65 Over and over again, we find deregulatory doctrinal moves coupled with this story of persecuted Christian conservatives.  In other words, this Court sees the Jehovah’s witness schoolchildren of Barnette everywhere, compelled against conscience to say the Pledge of Allegiance and salute the flag (more on this later).66 The Court in Barnette famously held that “[i]f there is any fixed star in our constitutional constellation, it is that no official, high or petty, can prescribe what shall be orthodox in politics, nationalism, religion, or other matters of opinion or force citizens to confess by word or act their faith therein.”67 But in recent cases, the Court sees this plight being faced by pharmaceutical marketers, healthcare providers, and commercial website designers. And the restrictions on “speech” or “compelled speech” at issue are not dogmatic but regulatory in nature.  Thus the paradigmatic oppressed speaker narrative unites the religious right and the deregulatory right of the current Court’s supermajority. From deacons to data processors—from gods to Google, indeed.  B.  The Expansion of First Amendment Coverage and its Connection to “Negative Theory” and the Oppressed Speaker Paradigm  Alongside a growing skepticism about the validity of government interests underlying a challenged regulation, First Amendment doctrine reflects a growing reach.  Freedom of expression today not only includes political speech traditionally understood—“I’m for Haley!” or “Vote for me!”—but also vulgarity, hate speech, commercial speech, falsehoods, interactive violent videos sold to minors, and threatening speech that falls short of the Court’s narrow definition of a true threat.68 Even nonsense is potentially protected speech,69 and protected speakers may include  62 Reed v. Town of Gilbert, 576 U.S. 155 (2015). 63 City of Austin v. Reagan National Advertising of Austin, LLC, 596 U.S. __, 142 S. Ct. 1464 (2022)(holding that a sign code that distinguished between on-premise and off-premise electronic signs was facially content-neutral regulation, and thus triggered only intermediate review, over Justice Thomas’s dissent claiming that this was inconsistent with Reed). 64 142 S. Ct. 1464 (2022). 65 NIFLA v. Becerra, 585 U.S. ___, 138 S. Ct. 2361 (2018).  66 West Va. State Bd. of Educ. v. Barnette, 319 U.S. 624 (1943). 67 Id. at 642. 68 See, e.g.,  Cohen v. California, 403 U.S. 15 (1971)(protecting vulgarity);  Central Hudson Gas & Electric Corp. V Public Service Commission, 447 U.S. 557 (1980)(setting forth test for protection of commercial speech); United States v. Alvarez, 567 U.S. 709 (2012)(noting that falsehoods not a per se categorical exception fo First Amendment coverage); Brown v Entertainment Merchants Association, 564 U.S. 786 (2011)(striking down California law that banned sale of violent videos to minors without parental permission); Counterman v. Colorado,600 U.S. 66 (2023)(setting forth revised test for ”true threats” that requires that defendant have some subjective understanding that statements are threatening to targeted person). 69  MARK V. TUSHNET, ALAN K. CHEN AND JOSEPH BLOCHER, FREE SPEECH BEYOND Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        17 17nonhumans that produce, edit, or merely convey information of value to listeners.70 Expression also need not be pure speech, but also may include expressive conduct71 and speech-facilitating conduct,72 posing hurdles to regulations aimed at non-speech harms.   As two of us have written elsewhere with Helen Norton, First Amendment coverage is ballooning, with vast implications for technology policy, such that regulation of much AI-generated speech may be covered by the First Amendment.73 The Court has found in recent years that money is “speech,” that data can be “speech,” and likely will find that content moderation is “speech,” for purposes of triggering First Amendment analysis. Even securities regulation, once squarely off the table, is now subject to First Amendment challenges and often scrutiny.74 This expanding coverage brings First Amendment analysis to bear on a growing number of regulatory areas. As a theoretical matter, such expansive coverage can be justified by multiple theories of the First Amendment—the marketplace of ideas, democratic deliberation, speaker autonomy—even when there is no human speaker, because of the human listeners involved.75 But in practice Court’s deregulatory moves are often justified by its oppressed speaker paradigm. Take the infamous case of Sorrell v. IMS, regarding a data privacy law (of sorts) passed by the Vermont legislature.76 The state law regulated the distribution of doctor-identifying information to pharmaceutical marketers.77 The Court reasoned that “the creation and dissemination of information are speech within the meaning of the First Amendment… Facts, after all, are the beginning point for much of the speech that is most essential to advance human knowledge and to conduct human affairs. There is thus a strong argument that prescriber-identifying information is speech for First Amendment purposes.”78 Despite this broad dicta, the Court did not hold that all data  WORDS: THE SURPRISING REACH OF THE FIRST AMENDMENT (2017) (discussing the reach of the First Amendment to include instrumental music, nonrepresentational art and “nonsense”). 70 See Massaro, Norton and Kaminski, supra note 13. 71 Id.  72 Jud Campbell, Speech Facilitating Conduct, 68 STAN. L. REV. 1 (2016) (explaining how doctrine protects some non-expressive conduct where it is essential to freedom of speech).  73 Massaro, Norton and Kaminski, Siri-ously 2.0., supra note 13. 74 Helen Norton, What 21st-Century Free Speech Law Means for Securities Regulation, 99 NOTRE DAME L. REV. 97 (2023); Michael R. Siebecker, Corporate Speech, Securities Regulation, and an Institutional Approach to the First Amendment, 48 WM. & MARY L. REV. 613 (2006); Frederick Schauer, The Boundaries of the First Amendment: A Preliminary Exploration of Constitutional Salience, 117 HARV. L. REV. 1765 (2004). 75 Massaro and Norton, Siri-ously; Massaro, Norton and Kaminski, Siri-ously 2.0, supra note 13. 76 Sorrell v. IMS Health, Inc. 564 U.S. 552 (2011)(striking down Vermont law that prohibited the sale, disclosure, and use of pharmacy records that revealed the prescribing practices of individual doctors without their consent, on grounds that the law was content-specific and speaker-specific insofar as it applied to pharmaceutical and data mining companies, but not health care researchers).  77 Id.  78 Id. at 570. Elsewhere, the Court says an “individual's right to speak is implicated when information he or she possesses is subjected to “restraints on the way in which the information might be used” or disseminated.”  Id. at 568. Electronic copy available at: https://ssrn.com/abstract=474217918  FROM GODS TO GOOGLE [28-Feb-24     18 is speech; nor did it hold that all data privacy laws must be subject to heightened scrutiny. Rather, the Court zeroed in on how Vermont’s law “impose[d] a speaker- and content-based burden on protected expression,”79 that justified close scrutiny. In other words, Sorrell was as much about protecting disfavored speakers with disfavored viewpoints as it was about data privacy.80 The Court feared that the Vermont legislature disfavored not only pharmaceutical advertisers as speakers, but their particular pro-marketing viewpoints on prescription drugs.81 It viewed with great skepticism Vermont’s proffered explanations of the government interests, and painted a picture of a world in which (poor, powerless) pharmaceutical marketers were persecuted by unconstrained (liberal) lawmakers. In the Court’s words, “[t]he law on its face burdens disfavored speech by disfavored speakers.”82 Framed in this way, a regulation stands almost no chance of being upheld by the Roberts Court.  This expansion of First Amendment coverage, premised on a theory that leads with deep suspicion of government regulators and exaggerated concerns about speaker vulnerability based on their viewpoints and speaker identity—versus their activities and potential harms to others—would be less fantastically bad if it were accompanied by genuine attention to legitimate regulatory interests. But the Court’s review of government interests shows little regard for these concerns; indeed, it is often outright scornful.   At the same time as it has expanded First Amendment coverage in ways particularly threatening to technology laws, the Roberts Court has tied its own hands by refusing to create new categorical exceptions to First Amendment protection in areas where the government has asserted harm-based reasons to do so. It insists that any categorical exception must be traditional and historical, not newly-minted or free-floating.83 The approach to privacy laws has thus shifted from careful case-by-case analysis of the balance between speech and privacy,84 to near-certain coverage of data privacy laws after Sorrell, with no room for the creation of “new” categorical exceptions for data privacy or AI regulation. There is no doctrinally principled way for the Court to “curb its free speech enthusiasm” in the zone of speech technology as such, whenever regulation can be placed into a negative rights and speaker-protective frame.  Covered speech abounds. So do regulations that are speaker-identity-specific and content-specific.    79 Id. at 571. 80 Cf. Richards, (arguing that most data privacy laws are constitutional). 81 Sorrell, 564 U.S. at 563 (noting that “pharmacies may sell the information to private or academic researchers, see § 4631(e)(1), but not, for example, to pharmaceutical marketers…”). 82 Sorrell  564 U.S. at 564. 83 See United States v Stevens, 559 U.S. 460 (2010)(rejecting argument for new categorical exception for depictions of animal cruelty or other exceptions that are not traditional and historical). See also United States v. Alvarez, 567 U.S. 709 (2012)(rejecting argument that falsehoods, without more, are a categorical exception to free speech coverage);  Brown v. EMA, 564 U.S. 786 (2011)(rejecting categorical exception for commercial sale to minors of violent interactive videos). 84 See, e.g., Bartnicki v. Vopper, 532 U.S 514 (2001)(holding that where media  broadcast a taped conversation that was illegally obtained by a third party, the defendant was not civilly liable). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        19 19C.  The Court’s Skepticism of “Content-Based Distinctions” and Fears of Viewpoint and Speaker Persecution  The Court’s fears of viewpoint and speaker persecution—particularly religious speakers expressing religious messages—motivate and color its recent jurisprudence. These fears have contributed to substantive doctrinal changes leading to new kinds of First Amendment challenges of technology laws, following Sorrell and its progeny Reed v. Town of Gilbert.85   In Reed, the Court doubled down on the persecuted speaker narrative. Gilbert, Arizona enacted a sign ordinance that generally banned the public posting of signs but carved out several exceptions.86  Exceptions were made for “ideological signs,” “political signs,” and “Temporary Directional Signs Relating to a Qualifying Event.”87 Each of these categories was treated differently, with “ideological signs” treated the most favorably.  The Good News Community Church and its pastor, Clyde Reed, wished to post temporary signs advertising church services without complying with the details of the ordinance, and challenged the ordinance as content-based discrimination. Justice Thomas, writing for the majority, held that because the law was content-based on its face, it triggered strict scrutiny.88 Content-based, the Court held, means “a law applies to particular speech because of the topic discussed or the idea or message expressed.”89 Thomas clarified that only laws that are facially content-neutral require any additional inquiry as to government motive and justification.90  Justice Breyer, writing in concurrence, rejected this reasoning entirely. He noted the potentially wide consequences: many laws facially classify based on content under the Reed definition, and would fall to strict scrutiny were Reed widely applied.91 Justice Kagan noted in concurrence numerous laws, including the Federal Highway Beautification Act, that for various reasons permit content-based exceptions to a general ban on signs.92 Kagan wrote that applying Reed’s holding literally would mean that regulators would have to “either repeal the exemptions that allow for helpful signs on streets and sidewalks, or else lift their sign restrictions altogether and resign themselves to the resulting clutter.”93 As Kagan observed, content-based regulations have historically triggered strict scrutiny only when “there is ‘any realistic possibility that official suppression of ideas is afoot.’”94  So why did the majority depart from this, and find that facially content-based laws  85 576 U.S. 155 (2015). 86 See id. at 180 (Kagan,J., concurring)(offering examples of other ordinances that regulate signs based on content, including Federal Highway law). 87 Id. at 159-60. 88 Id. at 163. 89 Id. at 163. 90 Id. at 164. 91 Id. at 177-78, (Breyer, J., concurring) (noting that “Regulatory programs almost always require content discrimination. And to hold that such content discrimination triggers strict scrutiny is to write a recipe for judicial management of ordinary government regulatory activity” and listing examples, including ”governmental regulation of securities…;of energy conservation labeling-practices…;of prescription drugs… and more)”. 92 Id. at 180 (Kagan, J., concurring). 93 Id. at 181. See also Sorrell (Breyer, J., dissenting). 94 Reed, 576 U.S. at 182. Electronic copy available at: https://ssrn.com/abstract=474217920  FROM GODS TO GOOGLE [28-Feb-24     20 always trigger strict scrutiny? Because of a fear about the persecution of minority religious speakers. Justice Thomas wrote that “one could easily imagine a Sign Code compliance manager who disliked the Church's substantive teachings deploying the [content based] Sign Code to make it more difficult for the Church to inform the public of the location of its services.”95 That is, a content-based law, even if it is not viewpoint-based or motivated by official suppression of ideas, gives too much leeway to enforcers to go after substantive messages and viewpoints—and disfavored religious actors.  We also see this paradigmatic story at play in the Court’s discussion in Reed of so-called “speaker-based” regulation and its relationship to content-based discrimination.96 Recall that in Sorrell the Court was concerned that Vermont’s law was viewpoint-based, a particularly egregious form of content-based discrimination that takes sides in a debate.97 But the Court in Sorrell also referred to “disfavored speakers.”98 Is it the case then, as some regulated technology companies appear to believe, that “speaker-based” discrimination also triggers heightened scrutiny? In Reed, the Court clarified that merely naming particular categories of speakers does not automatically trigger strict scrutiny.99 However, “laws favoring some speakers over others demand strict scrutiny when the legislature's speaker preference reflects a content preference.”100    Reed thus tightened the connection between “speaker-based” regulation and “content-based” regulation, potentially applying very demanding scrutiny to both. Along with the Court’s disdain for compelled speech,101 these shifts make it considerably harder for lawmakers to write laws that categorize the entities they regulate, or that compel specific disclosures, without triggering fatal-in-fact strict scrutiny.102 Impermissible speech compulsion is becoming coterminous with any regulation that aims at conduct but may compel disclosures, even if the government asserts compelling reasons for regulating that conduct. We are all Jehovah’s Witnesses now. It is no surprise that challenges to technology laws reflect these developments and  95Id. at 167-68 (“Accordingly, we have repeatedly ‘rejected the argument that ‘discriminatory ... treatment is suspect under the First Amendment only when the legislature intends to suppress certain ideas.’”). 96 “Speaker-based” discrimination has never been a separate category for First Amendment purposes. See discussion in NetChoice LLC v. Att. Gen, Fla., 34 F. 4th 1196 (11th Cir. 2022), cert. granted in part sub nom., Moody v. NetChoice, LLC, No. 22-277, 2023 WL 6319654 (U.S. Sept. 29, 2023) and cert. denied sub nom. NetChoice v. Moody, No. 22-393, 2023 WL 6377782 (U.S. Oct. 2, 2023). 97 Sorrell, 564 U.S. at 564. See also Reed, 576 U.S. at 168 (“Government discrimination among viewpoints—or the regulation of speech based on “the specific motivating ideology or the opinion or perspective of the speaker”—is a “more blatant” and “egregious form of content discrimination.”). 98 Sorrell , 564 U.S. at 564. 99 Reed , 576 U.S. at 170.  100 Id.  at 170 (stating that “a content-based law that restricted the political speech of all corporations would not become content neutral just because it singled out corporations as a class of speakers”). 101 See text accompanying notes 120-32 infra. 102 See Reed , 576 U.S. ___(Breyer, J., concurring) (providing examples of disclosures in securities regulation, privacy, elsewhere, that would trigger strict scrutiny as content-based. See also Genevieve Lakier, Reed v. Town of Gilbert, Arizona and the Rise of the Anti-Classificatory First Amendment, 2016 SUP. CT. REV. 233 (2017). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        21 21their worrisome impacts. A group of Internet Service Providers challenged Maine’s regulation of telecom privacy as being unconstitutionally speaker-based because it targeted service providers but not websites;103 and a California judge felt “troubled by [California’s children’s privacy law’s] clear targeting of certain speakers—i.e., a segment of for-profit entities, but not governmental or non-profit entities.”104  While the Court rolled back some of its reasoning in 2022 in City of Austin v. Reagan National Advertising of Austin,105 hurdles remain, especially for regulators’ attempts to increase transparency by requiring disclosures. In City of Austin, Justice Sotomayor, writing for the majority, explained that Reed could not be read to stand for the “extreme” principle that a signage law is content-based, and thus triggers strict scrutiny, if you have to read a sign in order to apply it.106 (Tellingly, Justice Thomas, who authored Reed, explained in his dissent that this was in fact what he had intended: “any law that draws distinctions based on communicative content is content based.”107) The Court in City of Austin held that laws distinguishing between “on-premises” and “off-premises” signs were not content-based regulation, but rather location-based regulation, despite the fact that an enforcer technically has to read a sign to determine whether it refers to on- or off-premises activity. Laws governing billboards have long made distinctions between signs referring to activity taking place on-premises versus off-premises. The Court reasoned that the doctrine consistently has held that “restrictions on speech may require some evaluation of the speech and nonetheless remain content neutral,” citing precedents allowing the regulation of solicitation as examples.108 Reed should be understood, per the majority in City of Austin, as finding that regulations are content-based if they discriminate based on “the topic discussed or the idea or message expressed.”109 In other words, the real problem in Reed was not that the city attempted to regulate signs announcing a one-time event—it was that the city attempted to regulate signs announcing events differently depending on whether they were “ideological,” “political,” or other (in that case, religious) in nature.110 This is an important correction, but City of Austin still leaves in place plenty of room for post-Reed problems and shenanigans, as Justice Breyer’s concurrence notes. The law in City of Austin may have been saved because of its specific reference to location, an element long found to be a neutral way of regulating speech under the time, place, and manner test.111 But as Breyer explains, “ordinary regulatory programs” such as security-related disclosures and laws governing  103 ACA Connects v. Frey, 471 F. Supp. 3d (318) (2020) (brief for Maine AG). 104  CAADCA NetChoice case at 15. 105 City of Austin v. Reagan National Advertising of Austin, LLC, 596 U.S. __, 142 S. Ct. 1464 (2022)(holding that distinctions between on-premise and off-premise electronic signs was facially content-neutral regulation, and thus triggered only intermediate review). 106  142 S. Ct.  at 1473-74. 107 Id. at 1481-82 (Thomas, J., dissenting)(arguing that the majority’s “skepticism is misplaced. We have often acknowledged that the need to examine the content of a message is a strong indicator that a speech regulation is content based”). 108 Id. at 1473-74. 109 Id. at 1471 (quoting Reed, 576 U.S. at 163). 110 Id. at 1475, n. 6. 111 Cantwell v. Connecticut, 310 U.S. 296, 307 (1940) (upholding regulation of the time, place, and manner of solicitation). Electronic copy available at: https://ssrn.com/abstract=474217922  FROM GODS TO GOOGLE [28-Feb-24     22 the labeling of prescription drugs are still likely to be found content-based even after City of Austin.112  For those justices who most consistently see a threat of religious persecution, the problem with City of Austin is that it fails to provide sufficient protection.  Justice Thomas’s dissent in City of Austin, joined by Justices Barrett and Gorsuch, once again clearly echoes the paradigm. Insisting that the majority is getting Reed wrong, Justice Thomas cites the importance of hand-tying the judiciary to a rigid test that would prevent judicial political preferences from winning the day.113 Once again, Thomas’s concern is for the minority religious speaker. He quotes his opinion in Reed, stating that a broad definition of content-based laws is necessary to prevent discrimination against religious speakers.114  To reiterate his view that distinctions between off-premises and on-premises signs are inherently content-based, he uses another example involving religious speech:  Take, for instance, a sign outside a Catholic bookstore. If the sign says, “Visit the Holy Land,” it is likely an off-premises sign because it conveys a message directing people elsewhere (unless the name of the bookstore is “Holy Land Books”). But if the sign instead says, “Buy More Books,” it is likely a permissible on-premises sign (unless the sign also contains the address of another bookstore across town). Finally, suppose the sign says, “Go to Confession.” After examining the sign’s mes-sage, an official would need to inquire whether a priest ever hears confessions at that location. If one does, the sign could convey a permissible “on-premises” message. If not, the sign conveys an impermissible off-premises message…115  The dissent’s priority is protecting not just any minority speech but religious minority speech. This is made all the more evident by the fact that Thomas dedicates three pages of the dissent to analysis of Hill, a case nowhere cited by the majority.116 Justice Thomas’s dissent reads the majority in City of Austin as implicitly relying on the analysis in Hill v. Colorado, an abortion clinic buffer-zone case.117 Thomas quotes Justice Scalia’s observation that “I have no doubt that this regulation would be deemed content based in an instant if the case before us involved antiwar protesters, or union members seeking to ‘educate’ the public about the reasons for their strike.”118 And just as the law in Hill could be misused to disfavor particular religious speakers, the law in City of Austin could, per Thomas, be misused to disfavor “signs conveying messages like ‘God Loves You’” and “dozens of religious and political messages that would be next to impossible to  112 142 S. Ct. at 1476 (Breyer, J., concurring). 113 Id. at 1481 (Thomas, J., dissenting) 114 Id. at1483 ( quoting Reed at 167-68) (“one could easily imagine a Sign Code compliance manager who disliked [a] Church’s substantive teachings deploying the Sign Code to make it more difficult for the Church to inform the public of the location of its services”). 115 Id. at 1481. 116 Id. at 1475(noting that Thomas’s dissent accuses the majority of resuscitating “a decision that we do not cite”). 117 Id. at 1490-92 (noting that “there is only one case that could possibly validate the majority’s aberrant analysis: Hill v. Colorado. That Hill is the majority’s only support underscores the danger”). 118 Id. at 1492. Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        23 23categorize under Austin’s rule.”119  Religious and political—in that order. Reed was not just a response to concerns about the persecution of minority political speakers. It was, per Thomas’s dissent in City of Austin, a response to Hill, where political liberals persecuted religious speakers.  D.  Compelled Speech and Crises of Conscience  Technology regulation is particularly at odds with this Court’s exceptional disdain for compelled speech in almost all forms.120 The Court relies heavily on the minority religious speaker paradigm to support its dire claims of regulatory persecution, but the doctrinal principles it has developed in this realm will not be so easily constrained.  A particularly telling recent example is National Institute of Family and Life Advocates v. Becerra, in which the Court held that California impermissibly compelled the speech of crisis pregnancy centers supported by NIFLA—a faith-based nonprofit that advances pro-life causes.121 The California law required that licensed clinics post notices alerting clients that the state provides free or low-cost family-planning services—including abortions—and providing a phone number that clients can call for further information about those services. It also required unlicensed clinics to post a notice alerting clients that the clinics’ personnel are not licensed to provide medical services.122  The Court held that the notices were a form of compelled speech that impermissibly burdened the clinics’ right to speak, were unrelated to the work of the pro-life clinics insofar as they did not perform abortion services, and distorted their pro-life messages in a viewpoint-specific and speaker-sensitive manner.123 The notices  119 Id. (emphasis added). 120 Many scholars have expressed concerns about the Court’s compelled speech doctrine --including its internal contradictions, its threat to valid government interests in requiring disclosures, and overclaims about when and how it genuinely, let alone unduly, burdens speaker autonomy. See, e.g., Jonathan H. Adler, Compelled Commercial Speech and the Consumer “Right to Know”, 58 ARIZ. L. REV. 421 (2016); Vikram David Amar & Alan Brownstein, Toward a More Explicit, Independent, Consistent and Nuanced Compelled Speech Doctrine, 2020 ILL. L. REV. 1, 6; William Baude & Eugene Volokh, Compelled Subsidies and the First Amendment, 132 HARV. L. REV. 171 (2018); Erwin Chemerinsky & Michele Goodwin, Constitutional Gerrymandering Against Abortion Rights: NIFLA v. Becerra, 94. N.Y.U. L. REV. 61 (2019); Alan K. Chen, Compelled Speech and the Regulatory State, 97 IND. L.J. 881 (2022); Caroline Mala Corbin, Compelled Disclosures, 65 ALA. L. REV. 1277 (2014); Lauren Fowler, The “Uncontroversial” Controversy in Compelled Commercial Disclosures, 87 FORDHAM L. REV. 1651 (2019); Abner S. Greene, “Not in my Name” Claims of Constitutional Right, 98 B.U. L. REV. 1475 (2018); Sarah C. Haan, The PostTruth First Amendment, 94 IND. L. J. 1351 (2019); Toni M. Massaro, Tread on Me!, 17 U. PA. J. CONST. L. 365, 404-15 (2014); Robert Post, NIFLA and the Construction of the Compelled Speech Doctrine,97 IND. L.J. 1071(2022) (discussing the confusion and its implications for a range of legitimate government regulations); Robert Post, Compelled Commercial Speech, 117 W. VA. L. REV. 867 (2015); Note, The Curious Relationship Between the Compelled Speech and Government Speech Doctrines, 117 HARV. L. REV. 2411 (2004); Martin H. Redish, Compelled Commercial Speech and the First Amendment, 94 NOTRE DAME L. REV. 1749, 1750 (2019); Seana Valentine Shiffrin, Compelled Speech and the Irrelevance of Controversy, 47 PEPP. L. REV. 731, 735-36 (2020); Steven H. Shiffrin, What is Wrong with Compelled Speech?¸ 29 J.L. & POL. 499 (2014); Eugene Volokh, The Law of Compelled Speech, 97 TEX. L. REV. 355, 357 (2018). 121 585 U.S. ___, 138 S. Ct. 2361 (2018).  122 138 S. Ct. at 2368-70. 123 Id. at 2371-78. Electronic copy available at: https://ssrn.com/abstract=474217924  FROM GODS TO GOOGLE [28-Feb-24     24 needed to be posted at the entrance of the facilities and in at least one waiting area, and to be at least 8 ½ by 11 inches, in no less than 48-point type.124  The Court refused to view the notices as a permissible form of professional speech regulation of the sort that has long endeavored to protect the interests of potential clients. Indeed, the majority noted that in other regimes, such as that of Nicolae Ceausescu and Nazi Germany, governments mandated that physicians give state-sanctioned advice to patients, which distorted medical independence and judgment.125 It expressed concern that American governments too might manipulate professional speech “to increase state power and suppress minorities.”126 To see this kind of threat required the Court to view the notices entirely as an abstraction, divorced from any practical assessment of the degree of incursion into matters of conscience actually inflicted by the compelled disclosures.  The notices were aimed at alerting patrons of these family planning centers of other options available to them, regarding a then-constitutionally protected right, about which the Court has held that states may weigh in more directly to control private and physician speech.127 Nothing mandated that these clinics or their personnel intone the government service messages, endorse them, point their clients to them, or refrain from criticizing abortion sharply as a violation of the sanctity of human life. But the Court nonetheless treated the notice requirement as a totalizing capture of the clinics’ expressive autonomy.  As we will explain, the failure to gauge the degree of speaker sovereignty left undisturbed by the challenged regulation, treating the complainant’s objection as the law’s only meaningful feature, was an error the Court would replicate in its most recent religious speech case, 303 Creative.128  The implications of Becerra for technology laws are significant, given the heavy reliance these laws place on mandated disclosures. Take, for example, a judge’s observation that the new California children’s privacy law had “[s]everal sections [that] require businesses to affirmatively provide information to users, and by requiring speech necessarily regulate it.”129 The law’s so-called “compelled speech” provisions included its requirement that businesses “[p]rovide any privacy information . . . concisely, prominently, and using clear language suited to the age of children likely to access that online service, product, or feature;”130 the requirement that businesses “provide an obvious signal to [a] child” if the child is being tracked or monitored by a parent or guardian via an online service, product, or feature;131 and the requirement that businesses “[p]rovide prominent, accessible, and responsive tools to help children . . .  124 Id. at 2370. 125 Id. at 2374 (quoting Berg, Toward a First Amendment Theory of Doctor-Patient Discourse and the Right to Receive Unbiased Medical Advice, 74 B.U.L. REV. 201, 201-02 (1994)(footnotes omitted). 126 Id. at 2374-75. 127 See Planned Parenthood Pa. v. Casey, 505 U.S. 833, 881-883 (1992)(Joint Opinion of O’Connor, Kennedy, Souter, JJ.)(stating it was not an undue burden on a person seeking an abortion to require the doctor to inform the person about the nature of the procedure, the likely gestational age of the fetus, the health risks of abortion and childbirth, the availability of printed materials describing the fetus, medical assistance for childbirth, potential child support and agencies that would provide adoption or other alternatives to abortion). 128 See infra text accompanying notes 133-67. 129 NetChoice CAADCA case: Case No. 22-cv-08861-BLF, 9/18/23. 130 CAADCA § 31(a)(7). 131 Id. § 31(a)(8). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        25 25exercise their privacy rights and report concerns.”132 The court then went on to find most of these requirements unconstitutional under the First Amendment.  A form of mandatory transparency that is central to the regulation of AI systems in other countries—individualized explanations of decision-making—is now up for evaluation at the Supreme Court.  Becerra itself would have been a formidable obstacle to the survival of these laws.  But as we will see, the doctrinal moves made by the majority in 303 Creative create unprecedented additional challenges for the vitality of technology regulation.  The first of these moves, as we explain in the next Part, was the Court’s decision to use free speech rather than free exercise principles, thereby developing principles that are not confined to the subset of cases where the speaker’s religious beliefs are at issue.    III. 303 CREATIVE: FROM FREE EXERCISE TO FREE EXPRESSION  303 Creative v. Elenis,133 the Court’s most recent religious speech case, involved a challenge to the Colorado Anti-Discrimination Act brought by a Christian website designer.  303 Creative posed a familiar dilemma: a person whose religious beliefs conflict with the requirements of a law that applies to them. It was a particular variant of this recurring conflict in that the conduct that plaintiff Lorie Smith wished to engage in—providing wedding websites on a commercial basis to opposite-sex couples but not to same-sex couples—was both religious and expressive, thus assertedly falling within the protection of both the Free Exercise and the Free Speech Clauses. Such circumstances give the Court the opportunity to decide which doctrinal principles to use.   In Masterpiece Cakeshop v. Colorado Civil Rights Commission, a previous case involving a Christian baker’s challenge to the same Colorado law, the Court squarely applied a free exercise analysis and sidestepped whether a wedding cake counts as speech.134 The Court relied upon comments of a Colorado Civil Rights Commissioner in concluding that the baker was subjected to anti-religious bias rather than the neutrality required by the free exercise clause.135  In 303 Creative, however, the Court clearly committed to a free speech analysis early on, agreeing to review Lorie Smith’s free speech claims but not her free exercise ones.  This choice is worth some reflection given its profound significance for the future of technology regulation.136  Kenji Yoshino argues that “the 303 Creative Court pivoted to free speech because the Court had reached an impasse in its free exercise jurisprudence.”137 The impasse arises from the Court’s growing disfavor for the framework set forth by Justice Scalia  132 Id. § 31(a)(10). 133 303 Creative, LLC v. Elenis, 600 U.S. 570, 143 S. Ct. 2298(2023). See Parts III and IV, infra. 134 Masterpiece Cakeshop, Ltd. v. Colorado C.R. Comm'n, 138 S. Ct. 1719, 1723, 201 L. Ed. 2d 35 (2018).  135 There is an extensive literature critiquing this conclusion, especially when viewed in light of the Court’s refusal to acknowledge the anti-Muslim animus of the Trump travel bans. For a partial sampling, see Leslie Kendrick & Micah Schwartzman, The Etiquette of Animus, 132 HARV. L REV. 133, 164 (2018); Mark Spindelman, Masterpiece Cakeshop’s Homiletics, 68 CLEVELAND ST. L. REV. 347 (2020). 136 Kenji Yoshino, Rights of First Refusal, 137 HARV. L. REV. 244, __ (2023).  137 Id. at ___. Electronic copy available at: https://ssrn.com/abstract=474217926  FROM GODS TO GOOGLE [28-Feb-24     26 in Smith v. Employment Division, in which the Court held that the Free Exercise Clause allowed states to prohibit peyote use, even for sacramental purposes, and thus states could deny unemployment benefits to someone discharged for religiously inspired peyote use.138 Distancing itself from earlier cases that were more protective of religious exemption rights, the Court held that “laws incidentally burdening religion are ordinarily not subject to strict scrutiny under the Free Exercise Clause so long as they are neutral and generally applicable.”139  This unpopular test has never been formally overruled, in part because some of the justices who are ready to see it go have openly expressed uncertainty about what should replace it.140 In that sense Yoshino is quite right to describe an impasse.141 At the same time, Smith has proven capacious enough to provide considerable protection to religious objectors.142   A law that has a single secular exemption – even if merely hypothetical – but none for religious reasons will not be considered generally applicable and will thus be subject to strict scrutiny.143 Using this approach in Fulton v. City of Philadelphia, the  138 Emp. Div., Dep't of Hum. Res. of Oregon v. Smith, 494 U.S. 872, 875 (1990). For a detailed chronicle of the “doctrinal dissembling” in Free Exercise jurisprudence that both pre and post-dates Smith, see James M. Oleske, Jr., Free Exercise (Dis)honesty, 2019 WIS. L. REV. 689, 690 (2019). 139 Fulton v. City of Philadelphia, Pennsylvania, 141 S. Ct. 1868, 1876 (2021). Fulton limited to the point of oblivion two prior cases in which the Court had applied strict scrutiny to determine whether a state was required to provide a religious exemption. See, e.g., Church of the Lukumi Babalu v. City of Hialeah, 508 U.S. 520 (1993); Masterpiece Cakeshop, Ltd. v. Colorado Civil Rights Commission, 584 U.S. ___, 138 S. Ct. 1719 (2018). 140 Fulton, 141 S. Ct. 1868, 1882 (2021) (Barrett, J., concurring). See also Christopher C. Lund, Answers to Fulton's Questions, 108 IOWA L. REV. 2075, 2077 (2023) (“Why should generally applicable regulations affecting religion categorically get strict scrutiny when many generally applicable regulations affecting speech do not?” and considering more generally differences between free exercise and free speech; conceptualizing freedom of religion “as a right of conscience and freedom of speech as something else.”).  141 See Zalman Rothschild, Free Exercise Partisanship, 107 CORNELL L. REV. 1067, 1094 (2022) (explaining how Smith has allowed for “an incredibly expansive view of religious discrimination to achieve not only what pre-Smith free exercise doctrine would have accomplished, but more.”). 142 See, e.g., Kennedy v. Bremerton, 142 S. Ct. 2427 (2022); Fulton, 141 S. Ct. 1868 (2021); Tandon v. Newsom, 141 S. Ct. 1294 (2021) (per curiam); Roman Catholic Diocese of Brooklyn v. Cuomo, 141 S. Ct. 63 (2020) (per curiam); Masterpiece Cakeshop, Ltd. v. Colorado Civil Rights Commission, 138 S. Ct. 1719 (2018); Church of Lukumi Babalu Aye, Inc., v. City of Hialeah, 508 U.S. 520 (1993). In all of these cases, the Court closely scrutinized the measures and found that each was either not “generally applicable,” or not neutral. Thus, strict scrutiny was triggered, under which test all of the measures failed.  143 As formulated by one of the theory’s most influential proponents, “The constitutional right to free exercise of religion is a right to be treated like the most favored analogous secular conduct.” Douglas Laycock & Steven T. Collis, Generally Applicable Law and the Free Exercise of Religion, 95 NEB. L. REV. 1, 22–23 (2016).  See also Douglas Laycock, The Remnants of Free Exercise, 1990 SUP. CT. REV. 1, 49 (arguing that “religion get something analogous to most-favored nation status. Religious speech should be treated as well as political speech, religious land uses should be treated as well as any other land use of comparable intensity, and so forth.”). The “single secular exemption” rule has been subject to withering criticism. See Andrew Koppelman, The Increasingly Dangerous Variants of the "Most-Favored-Nation" Theory of Religious Liberty, 108 IOWA L. REV. 2237, 2248 (2023) (“The laws against speeding make exceptions for ambulances.”). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        27 27Court held that Philadelphia was required to exempt religious foster care agencies from compliance with anti-discrimination provisions.  These provisions were not treated as generally applicable because there was a formal mechanism for granting exceptions, even though no exception had ever been actually given.144  The non-discrimination provisions were “therefore examined under the strictest scrutiny,” and failed.145  Fulton revealed that what the Justices dislike most about Smith—its lax treatment of neutral and generally applicable laws—can be narrowed to a point of virtual irrelevance without ever overruling Smith or having to choose what comes next.  This narrowing accelerated during recent litigation over Covid restrictions. The Court struck down public health officials’ efforts to contain the pandemic by restricting large gatherings on the ground that they selectively burdened religious worshipers’ desire to congregate, while allowing arguably similar health-imperiling gatherings.146 Especially given the exigent circumstances, the Covid cases dramatically showed what a thin channel government actors now must travel: “government regulations are not neutral and generally applicable, and therefore trigger strict scrutiny under the Free Exercise Clause, whenever they treat any comparable secular activity more favorably than religious exercise.”147  Free exercise, as this short summary reveals, is alive and well, notwithstanding the Court’s apparent hesitation over whether and how to overrule Smith. It is actively being sculpted into a doctrine that will frequently require the most searching review of government regulations, and the Court has not eschewed free exercise whenever a viable free speech claim allows it to cleanly sidestep the difficult Smith question.148 It is certainly true, as Yoshino observes, that in 303 Creative the religious exemption claims were “shunted over” to free speech jurisprudence,149 but to describe that as a result of the “impasse” over Smith fails to capture just how robust a free exercise framework was available to the Court given the work it had done to refashion Smith into a regime highly protective of religious objectors.  Only if the Colorado law had satisfied this increasingly demanding framework would the Court have needed to decide whether to overrule Smith in order to rule for the web designer. So perhaps the difficulty was that this would have been an uncomfortably close question: CADA presented the relatively rare case of a regulation with no exemptions of any kind, making it “generally applicable,” and there were no allegations in the record that it had been applied in a manner that would fail the  144 593 U.S. ___(2021).  145 Id. at ___. The Court construed the statutory and contractual provisions at issue in ways that allowed it to skirt broader issues about whether to overrule Smith, as some justices preferred.  146 See Tandon v. Newsom, 141 S. Ct. 1294 (2021) (per curiam); Roman Catholic Diocese of Brooklyn v. Cuomo, 141 S. Ct. 63 (2020) (per curiam); South Bay United Pentecostal Church v. Newsom, 141 S. Ct. 716, 716 (2021) (per curiam). 147 Tandon, 141 S. Ct. 1294, 1296, (2021) (emphasis added). See also Koppelman, supra note 25, at 2238 (explaining that the Court “now embraces what has been called by Laycock and others the “most-favored-nation” theory (hereinafter MFN), which holds that the denial of a religious exemption is presumptively unconstitutional if the state “treats some comparable secular activities more favorably.”). Cf. Nelson Tebbe, The Principle and Politics of Equal Value, 121 COLUM. L. REV. 2397, 2403 (2021) (considering whether “equal value is being administered in the service of a problematic political program.”) 148 Kennedy v. Bremerton School District, 597 U.S. __(2022).  149  Yoshino, supra note 136. Electronic copy available at: https://ssrn.com/abstract=474217928  FROM GODS TO GOOGLE [28-Feb-24     28 neutrality assessment.150 The Court conceivably could have adopted the reasoning of appellate Judge Tymkovich, who argued that “the entire CADA enforcement mechanism is structured to make case-by-case determinations” and thus was equivalent to the potential for individualized exemptions invalidated in Fulton.151 But this clever sleight of hand  treats the discretion that is inherent in any law enforcement regime as a means by which a government authority is “parceling out exceptions” based on secular reasons, and therefore required to do so for religious ones.152 Embracing this logic to rule against Colorado without overruling Smith would have been tantamount to declaring the category of “neutral and generally applicable laws” to be an empty set.  Whether the Court were to follow this rationale or instead had managed to finally overrule Smith explicitly, in either case the next step under Free Exercise doctrine would be the application of strict scrutiny: after all, that is what the pre-Smith regime of Sherbert v. Verner153 and Wisconsin v. Yoder154 had provided.  But under the free speech theory that the Court used instead, the Colorado law was struck down without even receiving a strict scrutiny analysis.   The Court’s pivot to speech principles produced a doctrinal framework that was even more protective of religious speakers than the most robust imaginable version of free exercise. And in its eagerness to ensure a victory for religious believers like Lorie Smith, unhampered by the principles that continue to constrain free exercise doctrine in spite of all the uncertainty, the Court used a doctrinal framework that is indifferent to whether the objections presented are based on sincere religious belief.155  Aside from the irony, that move takes a significant step from gods to Google.  Free exercise claimants make claims based on “the spiritual foundations of their deeply held beliefs and commitments.”156 The believer’s sincerity is an element of the analysis and courts must confirm that the challenged law places a “substantial burden” on the exercise of religion.157 That this is an obvious and uncontroversial feature of free exercise jurisprudence—perhaps the one thing that has remained constant over decades of fluctuation—should not obscure its significance, especially as contrasted with free  150 The allegedly anti-religious comments by the Colorado official that gave the Court an easy answer for ruling in favor of the religious cakemaker in Masterpiece Cakeshop had not recurred. The Court conceivably could have treated CADA as forever tainted by those past comments, but this would have been an astonishing break from its approach in a wide range of cases where the Court has refused to find a law invalidated by prior manifestations of discriminatory intent. 151 6 F.4th 1160, ___(10th Cir. 2021) (J. Tymkovich, dissenting). 152 Id.  153 374 U.S. 398 (1963). 154 406 U.S. 205 (1972). 155 Nathan S. Chapman, The Case for the Current Free Exercise Regime, 108 IOWA L. REV. 2115, 2122–23 (2023) (“Government agencies and courts adjudicating such claims must ensure that they are based on religion (or conscience, for claims based on accommodations extending to nonreligious conscience) and that the claimant's beliefs are sincere).  156 CHRISTOPHER L. EISGRUBER & LAWRENCE G. SAGER, RELIGIOUS FREEDOM AND THE CONSTITUTION 70 (2007).  157 See also Christopher C. Lund, Answers to Fulton's Questions, 108 IOWA L. REV. 2075, 2077 (2023) (conceptualizing freedom of religion “as a right of conscience and freedom of speech as something else.”). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        29 29speech doctrine.158 As Douglas Laycock has put it, a free speech theory protects “even simple bigots in any expressive context instead of just those with sincere religious objections in religious contexts.”159 Because free speech protection does not rest on sincere religious belief it may protect not only simple bigotry in the full range of human expression but also that of human-built technology: the algorithms and machines playing an ever-expanding role in daily life.160   By using free speech to relieve Smith of the obligation to comply with anti-discrimination law, the Court opened the door to all other similarly situated speakers—not only conscience-driven religious ones—to successfully assert compelled speech claims. The implications for information technology, where the powerful entities resisting regulation have no plausible claim to sincere religious belief but can readily assert that their products and services have some expressive dimension that is being “coerced,” should be clear.  The move from free exercise to free speech, moreover, shifts the terrain from what is essentially an individualized exemption system to a frontal attack on the very validity of the core regulatory enterprise. Free exercise doctrine presupposes that the laws to which religious believers are objecting are laws that state actors are free to enact and enforce as long as they do not unduly burden religious exercise.161 By choosing free speech grounds, the Court had no basis to speak in terms of a religious exemption from an otherwise valid law, and indeed it did not. Quite the contrary, the Court repeatedly expresses its profound disapproval of what it viewed as Colorado’s “effort to eliminate disfavored ideas.”162 It treated West Virginia State Board of Education v. Barnette163 as an easy and obvious governing precedent for its conclusion that individuals cannot “be  158 See Sherbert v. Verner, 374 U.S. 398 (1963) (requiring for the first time that unless a state can satisfy strict scrutiny it must accommodate those whose failure to comply with a law “is due to their religious convictions”). See also Yoshino, supra note 136 (“to briefly state the obvious, the religion clauses protect individuals only on the basis of their religion.”). 159 Douglas Laycock, The Broader Implications of Masterpiece Cakeshop, 2019 B.Y.U. L. REV. 167, 182 (2019). 160 To qualify for a religious accommodation, a claimant must demonstrate sincerity. Nathan S. Chapman, Adjudicating Religious Sincerity, 92 WASH. L. REV. 1185, 1187 (2017); Vikram David Amar & Alan E. Brownstein, Exploring the Meaning of and Problems with the Supreme Court's (Apparent) Adoption of a “Most Favored Nation” Approach To Protecting Religious Liberty Under the Free Exercise Clause: Part One in a Series, JUSTIA (Apr. 30, 2021), https://verdict.justia.com/2021/04/30/exploring-the-meaning-of-and-problems-with-the-supreme-courts-apparent-adoption-of-a-most-favored-nation-approach-to-protecting-religious-liberty-under-the-free-exercise-c [https://perma.cc/9869-WNP5] (on risk of sham claims). While scholars have found that within free exercise adjudication “sincerity has become a meaningless requirement,” it is still formally part of free exercise claims. Xiao Wang, Religion As Disobedience, 76 VAND. L. REV. 999, 1005 (2023). 161 See Douglas Laycock, The Broader Implications of Masterpiece Cakeshop, 2019 B.Y.U. L. REV. 167, 173 (2019)(arguing that Smith’s treatment of Sherbert, while characterized by some as a disingenuous retrofitting, speaks of an individualized decision-making process).  162 During oral argument Gorsuch angrily referred to re-education programs. https://slate.com/news-and-politics/2022/12/gorsuch-reeducation-discrimination-lgbtq-civil-rights-303-creative.html.  163 W. Virginia State Bd. of Educ. v. Barnette, 319 U.S. 624 (1943). Electronic copy available at: https://ssrn.com/abstract=474217930  FROM GODS TO GOOGLE [28-Feb-24     30 conscripted to disseminate the government’s preferred messages.”164 In Barnette, the Supreme Court famously found that Jehovah’s Witness schoolchildren have a free speech right not to recite the Pledge of Allegiance or salute the flag upon pain of expulsion from school combined with criminal sanctions for their parents.  To draw on Barnette the Court had to treat a public accommodations law, prohibiting discrimination against individuals on the basis of protected status, as the doctrinal equivalent of a compulsory “affirmation of patriotic loyalty.”165 CADA was not treated as a presumptively valid law subject to a religious exemption, but as a coercion of thought and expression that is intrinsically wrongful. This is an extraordinary refashioning of public accommodation law.  It would have been unnecessary under a free exercise framework, which would have focused on whether Colorado could provide an exemption to religious dissenters like Smith without unduly undermining its general regulatory purposes.166  The consequences of the Court’s choice to use free speech rather than free exercise principles in 303 Creative are now in full view. If the government is not merely required to grant individualized exemptions, but prohibited from anything describable as the conscripted dissemination of its preferred message, and the objection can be pressed by any and all speakers regardless of religious belief, then what began as a concern for religious freedom has a much vaster deregulatory reach.167 The Court’s reasoning cast doubt on whether secular expressive conduct in places of public accommodation can be regulated; it also potentially imperiled the vast array of technology regulation that applies to any compelled conduct that is as expressive as web designing.      Having chosen the free speech path, the Court should have navigated and applied the doctrine more cautiously, and faithfully. But as we explain in the next Part, the Court instead used a free speech analysis that seems to have been crafted specially for the benefit of religious believers but logically cannot be confined to them.   IV. SPEECH DOCTRINE CRAFTED FOR RELIGIOUS BELIEVERS  .  303 Creative threatens to make information technology impervious to regulation. It not only is part of the increasingly lethal free speech weapon available to litigants defending against government regulation that we have described above. It represents a particularly aggressive advance of this free speech revolution in several respects. In 303  164  600 U.S. at 584-85. 165 Id.  166 Even if Colorado had lost at some stage of the analysis, it could have simply been ordered to provide an exemption rather than being scolded for attempting to “coopt an individual’s voice for its own purposes,” id. at 592 and “trying to eliminate ideas that differ from its own.” Id. at 597. 167 See, e.g., Douglas Laycock, The Broader Implications of Masterpiece Cakeshop, 2019 B.Y.U. L. REV. 167, 173 (2019). See also 108 IOWA L. REV. 2237, 2276 (noting that “[r]eligious accommodations always involve a guess about whether there will be so many claims that the law's purpose will be thwarted--whether the exemption of the Catholic Mass from the 1919 Volstead Act's prohibition of alcohol would lead huge numbers to convert to Catholicism just so they can imbibe (it didn't), or whether exempting all pacifists would hamstring the military draft (at the end of the Vietnam War, it did.”).  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        31 31Creative, the Court concluded that refusing to design wedding websites for same sex couples was “pure speech” rather than a combination of conduct and speech. It concluded that the speech in question belonged to the vendor rather than the customers for whom the website was designed. It avoided any assessment of whether the compliance burden was unduly coercive. Having made each of these troubling moves in the website designer’s favor, contrary to ample precedent instructing otherwise, the Court then made the most troubling move of all: rather than apply strict scrutiny, it concluded that the law was categorically unconstitutional full stop, no matter the government interest.  In this Part, we analyze each of these significant steps in turn. But we note at the outset that it did not have to be this way. The Court could have declined to hear 303 Creative for a number of reasons. Lorie Smith had never before designed a wedding website, had never been approached by a same-sex couple seeking her services when her lawsuit was filed,168 and had never been punished under the nondiscrimination mandate, though Colorado “‘decline[d] to disavow future enforcement’ proceedings against her.”169 The pre-enforcement, essentially hypothetical posture of the case meant it proceeded without the clarifying impact of having actual same-sex couples who sought Smith’s services and denied them a seat at the litigation table. In concluding that the effect of a purportedly credible threat of enforcement was to impermissibly compel Smith’s speech and silence, the Court distorted precedent and expanded the deregulatory impact of the case well beyond the context of religious web designers.   A. “Pure Speech” and the Erosion of the Speech/Conduct Distinction  Free exercise doctrine is indifferent to the distinction between speech and conduct, but free speech protection is not. Free speech analysis draws a distinction between two kinds of claims: those where a government regulation directly infringes a claimant’s protected speech, and those where the government aims to regulate conduct and only burdens speech incidentally.170 The first kind of regulation typically is subject to strict scrutiny, at least where the regulations are content-based; the second receives only  168 See Victoria Bisset and Jaclyn Peiser, Man Cited in Supreme Court LGBTQ Rights Case Says He Was Never Involved, WASH POST, July 1, 2023. In court documents, Smith’s lawyers cited a query that they claimed was sent by an individual named “Stewart” with contact information that matched a person named Stewart who denied he had sent the inquiry. The alleged request asked for Smith’s services for Stewart’s forthcoming wedding to a person named “Mike” and stated: “We are getting married early next year and would love some design work done for our invites, placenames etc. We might also stretch to a website.” However, Stewart –a San Francisco man married to a woman --told reporters who contacted him that he had never contacted Smith. https://www.washingtonpost.com/politics/2023/07/01/supreme-court-colorado-website/?utm_campaign=wp_politics_am&utm_medium=email&utm_source=newsletter&wpisrc=nl_politics; Melissa Gira Grant, The Mysterious Case of the Fake Gay Marriage Website, the Real Straight Man, and the Supreme Court, The New Republic (June 29, 2023), available at: https://newrepublic.com/173987/mysterious-case-fake-gay-marriage-website-real-straight-man-supreme-court.  169 600 U.S. at 587-88(citing the Tenth Circuit opinion, which held that Smith had standing to sue, but ruled against her on the merits). 170 Texas v. Johnson, 491 U.S. 397 (1989) (holding flag burning may be free speech and was impermissibly prohibited under a federal law that made flag “desecration” a crime).  Electronic copy available at: https://ssrn.com/abstract=474217932  FROM GODS TO GOOGLE [28-Feb-24     32 intermediate scrutiny.171 The first question for the Court, then, was whether the Colorado law sought to regulate speech or conduct.  Justice Gorsuch treated the complex and doctrinally significant speech-conduct question as resolved by Colorado’s stipulation that the customized website content was expressive.172 This was a factually misleading move that dodged the key threshold question for free speech analysis: whether any impairment of Smith’s purported expression was an incidental byproduct of a law aimed at regulable conduct.173 It also did not resolve the ultimate question of whether the state’s regulation permissibly prohibited that expression given the appropriate standard of review for the regulation’s conduct-aimed purpose and design.  To understand why Colorado’s stipulation could not possibly perform all the analytical work that free speech doctrine requires in this case, it is useful to consider some contrasting examples.  Assume that a political protester on a streetcorner holds up a sign. This protester is engaged in protected speech, in a form and a place where free speech rights are strongest. Government efforts to sanction such speech based on its content must satisfy the most demanding form of judicial scrutiny, which the government rarely can survive.174 This is particularly so if the government regulation compels the regulated person to speak or remain silent.175 But consider the same speaker, who now intends to express the same political idea by detonating a bomb in a government building. The speaker announces this intention, to make sure that all clearly understand the meaning of the act. That the bombing is an expressive act does not place it under the free speech umbrella at all, let alone insulate the speaker from liability for the crime. This is doctrinally treated as pure conduct and can be regulated as such.  If the same speaker decides to bellow her political message through a bullhorn rather than displaying a sign, the government might permissibly regulate the noise, but not the message. Content-neutral regulation that incidentally affects protected speech can be analyzed under the Court’s “time, place and manner test,” outlined in Ward v. Rock Against Racism.176 So long as the government isn’t aiming the law at the content or message and leaves open “ample alternative channels of communication,” a law  171 United States v. O’Brien, 391 U.S. 367 (1968) (holding that a federal law that made destruction or mutilation of a draft card a crime could be applied against a person who burned his draft card to protest war).  172 600 U.S. at 587-88 (stating that it agreed with the Tenth Circuit that the wedding websites would be pure speech, and that this was a “conclusion that flows directly from the parties’ stipulations.”). He also read the state’s stipulations to concede that this was Smith’s speech, not the hypothetical wedding couple’s. Id. at 588.  173 What the state actually conceded was that all of Smith’s graphic designs and website designs were “expressive in nature, as they contain images, words, symbols, and other modes of expression that Plaintiffs use to communicate a particular message.” See Blake E. Reid, On Stipulations and Expressiveness, July 4, 2023, available at https://blakereid.org/on-stipulations-and-expressiveness/. 24145 Cover.pdf (supremecourt.gov). p. 181a. Stipulations 46 and 47. (quoting Merriam-Webster’s first definition of expressiveness). Observers who criticized Colorado’s stipulations may not have realized how much they were distorted by the Court’s characterizations.  174 See Reed, 576 U.S. 155 (2015) (defining facially content-based sign regulations that trigger strict scrutiny to include any regulation that hinges on what the sign says). Cf. City of Austin v. 142 S. Ct. 1464, see notes 106- 19 supra. 175 See, e.g., Becerra, 138 S. Ct. 2361 (2018)(). See text accompanying notes 121-28 supra. 176 Ward v. Rock Against Racism, 491 U.S. 781 (1989). Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        33 33governing the noise historically would be likely to be upheld.177 In general, when a speaker chooses a conduct-laced means of expression that includes actions that the government has an important interest in regulating–e.g. preventing property destruction, fires, littering, trespassing, defacing of sacred objects, physical harms to others, or disruptions of order—then the Court applies the United States v. O’Brien test.178 The test asks whether the government interest in regulating the conduct is sufficiently important, and whether the regulation of the conduct was aimed at the conduct, not the speech, and reaches no farther than necessary in doing so.179 The test does not turn on whether the regulation compels speech or silence: it focuses on whether the government has a legitimate interest in regulating the conduct, then proceeds to the question of whether its incidental effect on expression is justified. So, for example, imagine that a state passes a law making it a misdemeanor to purposely deface, damage, or otherwise physically mistreat a cemetery or gravesite without privilege to do so. The law is applied against a person who spray-paints swastikas on the gravesites of Jewish people. Painting a swastika is a form of political speech; defacing or damaging a person’s gravestone with paint is conduct. Punishing the person for this expressive act is both an infringement on speech and a means of punishing the especially heinous conduct of gravestone defacement. This regulation would fall into the constitutional grey zone between speech regulation and conduct regulation. The outcome would depend on a court’s weighing of the government interest versus impact on expression, through the O’Brien test. The web designer’s speech and conduct in 303 Creative also fell into this constitutional grey zone. Her refusal to design a wedding website for same-sex customers was most accurately characterized as a conduct-laced means of expressing her beliefs. The conduct in question—discrimination—was the true target of the government measure. Colorado sought to regulate the “refusal to serve” step of her expressive activities, and any announcement of her discriminatory intentions to do so on the site. It rendered impermissible the following conduct: refusing to provide wedding websites for commercial sale on equal terms regardless of sexual orientation.  Prohibitions on discriminatory conduct by businesses that hold themselves out to the public have been around since the 1960s180—with some versions of public accommodations and common carriage protections as ancient as Roman law.181  In keeping with this longstanding tradition of regulating commercial activity, CADA  177 For critique, see Enrique Armijo, The Ample Alternative Channels Flaw in First Amendment Doctrine, 73 WASH. & LEE L. REV. 1657 (2016). 178 391 U.S. 367 (1968). But see Lund, supra note 157, at 2082 (arguing that “O'Brien is not the rule in any of the compulsion cases”).  179 391 U.S. 367 (1968). 180 Title II, Civil Rights Act 1964, 42 U.S.C. sec. 2000. Precursors of the laws began much earlier, as part of Reconstruction, though their force was eviscerated by the Civil Rights Cases in 1883. 109 U.S. 3 (1883)(striking down federal legislation banning private discrimination in places of public accommodation, on ground that congressional power to enforce the equal protection clause and other provisions of the Fourteenth Amendment was limited to regulation of state government; the Court later held that Congress has power to regulate places of public accommodation under the Commerce Clause).  181 See, e.g., https://core.ac.uk/download/pdf/56353394.pdf on the duty of innkeepers to host passing travelers. Electronic copy available at: https://ssrn.com/abstract=474217934  FROM GODS TO GOOGLE [28-Feb-24     34 should have been viewed as a regulation of conduct that sweeps up otherwise protected expression, giving rise to O’Brien’s two-step analysis.182 The first step is about First Amendment coverage: it asks whether designing a web site for a same-sex wedding is speech or expressive activity covered by the First Amendment. The answer to this is most likely—though not certainly—“yes.” At the stage of web design, Smith would be using words, text, and artistic design. This will probably found by courts to be “speech” in the sense that it is covered under the increasingly expansive First Amendment,183 though we do not think this is the only or even best way to characterize web design regulation, especially when that regulation is aimed explicitly at conduct (for example, manipulation or fraud). Nevertheless, that customized web designs may be speech covered by the First Amendment is not dispositive. The second step of O’Brien asks if the speech is being infringed directly (e.g. compelling Smith to affirm or celebrate same-sex marriage) or is being infringed indirectly via a government regulation aimed at her conduct (e.g. requiring Smith to serve same-sex couples on equal basis with other customers when seeking the same services). If so, courts apply intermediate scrutiny to the regulation, not strict scrutiny. The question becomes whether Colorado could regulate the conduct component of her refusal to serve, without unduly or directly infringing upon the expressive component of that same conduct.  The 303 Creative opinion did not provide a satisfactory answer to either question, mentioning O’Brien only in a footnote.184 Instead, Justice Gorsuch dodged O’Brien’s two-step inquiry by characterizing Colorado’s regulatory intentions as directed at the compulsion of speech, rather than the prohibition of discriminatory conduct. In effect, Gorsuch noted, Colorado “seeks to compel just the sort of speech that it tacitly  182 But for a potentially problematic fissure, see Lund, supra note 157 at 2082. 183 Spence v. Washington set forth the starting point for defining what counts as speech under the First Amendment fifty years ago. The Court stated that when Mr. Spence displayed an American flag with a peace symbol affixed to it made out of removable tape, he was engaged in speech. 418 U.S. 405 (1974). His “message was direct, likely to be understood, and within the contours of the First Amendment.” Id. at 415. The Court further noted that the speaker had “[a]n intent to convey a particularized message… and in the surrounding circumstances the likelihood was great that the message would be understood by those who received it.” Id. at 410-11. The Court later stated, however, that First Amendment speech extends much farther. Assessing the speech status of a parade, the Court has held that a “narrow, succinctly articulable message is not a condition of constitutional protection,” noting that such a rule would fail to reach much protected speech such as a Jackson Pollock painting. Hurley, 515 U.S. at 569. Moreover, that a parade organizer combines “multifarious voices” into its expressive message does not mean the organizers lose constitutional protection. Id. There thus is no “particularized message” condition on speech protection. See JENNIFER PETERSEN, HOW MACHINES CAME TO SPEAK: MEDIA TECHNOLOGIES AND FREEDOM OF SPEECH 87-119 (2022) (discussing the relaxation of this aspect of speech coverage).  Nor is there a general requirement that others will reasonably perceive the speaker to be uttering a specific message. See Alan K. Chen, Cheap Speech Creation, 54 U.C. DAVIS L. REV. 2405, 2415-17 (2021) (describing speech-generative acts such as newsgathering that may not, taken alone, express a particular idea). Even gibberish may be protected speech. See Tushnet,  Chen and  Blocher, supra note 69 (discussing the reach of the First Amendment to include instrumental music, nonrepresentational art and “nonsense”). 184 600 U.S. at 600 fn.6.  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        35 35concedes lies beyond the reach of its powers” insofar as it acknowledged the speech component of Smith’s refusal to serve same-sex wedding design customers.185 But as we saw with the swastika on the gravestone example above, the deployment of an indisputably expressive message does not make conduct immune from regulation. Recognizing a speech component begins rather than ends the inquiry.  To characterize public accommodations law as primarily concerned with the production of state-favored speech rather than the protection of historically marginalized individuals also reflects a profound logical error. Colorado sought to enforce the same prohibition on discriminatory treatment against public accommodations that are incontrovertibly expressive (e.g. web designers) and public accommodations that are, at least for the present, treated as non-expressive (e.g. hotels and restaurants). If the regulation is one and the same, regardless of whether there’s speech in the mix, it cannot be said that Colorado seeks to compel speech itself, as opposed to prohibiting discriminatory conduct.  Colorado did not compel her to design a wedding website for a same-sex couple any more than it required her to crochet tea cozies – it merely set forth an equal treatment condition on the provision of her commercial services.  And it argued that it had a compelling reason to do so: protecting the dignity and market-access rights of historically marginalized people.  Instead of conducting a thorough inquiry into this second step, Justice Gorsuch mashed together passages from compelled speech cases that did not engage O’Brien and declared that government cannot force a person to “‘utter what is not in [her] mind’ about a question of political and religious significance.”186  To reinforce his conclusion that CADA regulates speech rather than conduct, Justice Gorsuch emphasized that Smith’s denial of service to a same-sex couple would not be based on their “status” (“No LGBTQ people served here”). Smith would serve gay customers, just not any seeking wedding web design services and messages.187 Indeed, she would not serve any customers—gay or straight—whose messages might violate her beliefs.188 But this attempted distinction ignores that discrimination cases have long entailed both conduct and viewpoints—e.g. racist ideologies, thoughts, motives, or biases that animate the conduct. An employer can fire an employee for no reason, but not for a discriminatory reason, despite how the latter implicates viewpoint-driven expression.189 When the discriminatory viewpoint is expressed through conduct, then it becomes an unlawful act—more like a physical block of an individual than like pure communication on a public streetcorner.190 As the Court recognized in Giboney v.  185 Id. at 587-88. 186 Id. at 596 (quoting, again, Barnette). Justice Gorsuch’s answer to O’Brien is simply to cite Hurley for the proposition that “there is nothing “incidental” about an infringement on speech when a public accommodations law is applied “peculiar[ly]” to compel expressive activity. 303 Creative, 600 U.S. at 600 n.6. But there was nothing peculiar about this application. See also Linda C. McClain, Do Public Accommodations Laws Compel “What Shall Be Orthodox?”, paper on file with authors (noting, as do we, how use of Barnette divorced from its mandatory flag salute context distorts the case and its proper applications). 187 Id. at 598. 188 Id.  189 See, e.g., Hishon v. King & Spalding, 467 U.S. 69, 78 (1984)(rejecting argument that Title VII violates employers’ First Amendment rights). 190 336 U.S. 490, 502 (1949). The Court thus has upheld nondiscrimination laws applied to Electronic copy available at: https://ssrn.com/abstract=474217936  FROM GODS TO GOOGLE [28-Feb-24     36 Empire Storage & Ice Co., “it has never been deemed an abridgement of freedom of speech or press to make a course of conduct illegal merely because the conduct was in part initiated, evidenced, or carried out by means of language, either spoken, written, or printed.”191 The Court in prior antidiscrimination cases landed on the “regulable conduct, despite the speech component” side of the speech-conduct continuum.192 303 Creative disrupts the careful balancing these precedents reflect.  The results not only may imperil anti-discrimination law, as others have noted,193 but will also bedevil technology regulation, where the speech/conduct distinction is particularly challenging. Information is everywhere when it comes to online platforms. And it is often intimately entwined with some kind of conduct. In the digital context, technologies often simultaneously communicate something and do something. Software that decrypts or encrypts also uses symbols and words.194 Should software be protected as speech?  Can it be regulated as a potentially dangerous product violating export controls? 195  Personal data that enables complex market manipulation can also be characterized as a necessary precursor to commercial speech.196 A website isn't just verbal content: it's simultaneously design, in the sense that it also consists of consent boxes that urge a user to click "yes," manipulative dark patterns, and even addictive elements—by design.197 On this understanding, the Federal Trade Commission (FTC)  public accommodations, when confronted with First Amendment challenges. See, e.g., Newman v Piggie Park Enters., Inc., 390 U.S. 400 (1968); Roberts v. U.S. Jaycees, 468 U.S. 609 (1984); Board of Directors of Rotary Int’l v. Rotary Club of Duarte, 481 U.S. 537 (1987); New York State Club Ass’n v. City of New York, 487 U.S. 1 (1995). Indeed, although Justice Gorsuch relies on Hurley, the Court there noted that public accommodations laws “do not, as a general matter, violate the First or Fourteenth Amendments.” 515 U.S. 557, 572 (1995). See also Wisconsin v. Mitchell, 508 U.S. 476 (1993)(unanimously holding in an opinion by Chief Justice Rehnquist that where a defendant’s sentence for aggravated battery was enhanced because he intentionally selected his victim because of his race, this did not violate the First Amendment). 191 336 U.S. 490, 502 (1949). 192 Central to these cases is the understanding that places of public accommodation are not streetcorners where speech rights are maximal. Think, in contrast, of the Westboro Baptist Church picketing a serviceman’s funeral with speech that the family of the deceased found not only hateful, but an intentional infliction of emotional distress. Snyder v. Phelps, 562 U.S. 443 (2011). This speech nevertheless was protected because it was on the streetcorner – a public forum – not at the soldier’s graveside, not in the workplace, not at the memorial service, or any other bounded place where the speech need not be tolerated and the private speaker’s silence could be “compelled.” Had the church opened a place of public accommodation, protection of this diatribe would have been a much harder case. As Jacob Eisler has argued, the protection of speech in places of public accommodation ignores the competing framing of these as sites entailing “a shared social practice that is legitimately shaped by collective political decisions.” Discrimination, Private Liberty, and Public Accommodations Law (paper on file with authors). 193 See, e.g., David Schwartz, Making Sense of 303 Creative: A Free Speech Solution in Search of a Problem, paper on file with authors. 194 Robert Post, Encryption Source Code and the First Amendment, 15 BERKELEY TECH. L. J. 713 (2000) (“From a constitutional perspective, it is one thing to use source code to convey ideas to an audience, and it is quite another to use source code to run a computer.”). 195 Id.  196 Ryan Calo, Digital Market Manipulation, supra note 42.  197 A multistate group of 32 Attorneys General filed suit against Meta against Meta for among other things falsely representing that its platform is “safe and not designed to induce Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        37 37has been targeting deceptive website design for decades.198 Effective regulation of the digital realm in an era of proliferating First Amendment litigation depends on courts being able to disentangle the communicative elements of technology from conduct, and maintain a space for legitimate government regulation of the latter. Finding that merely because online actions involve “images, words, symbols, and other modes of expression” they trigger First Amendment coverage and almost-automatic protection could make regulation nigh-impossible.   B. Conflating Customer Speech and Vendor Speech  Even assuming that the discriminatory provision of customized wedding websites is best considered speech rather than conduct, recall that Smith claimed to be designing websites for other people.  So do these websites constitute the speech of the designer or the customer who requests and pays for the product?  The Court in 303 Creative settled on the former with little analysis. It treated Smith as the relevant messenger in two respects. If the Colorado law were enforced against her, the Court reasoned, she would be required to design a customized wedding web site for a same sex couple. To do so she would “vet” each potential project, consult with clients to elicit their “unique stories as source material,” and “produce a final story for each couple using her own words and her own “original artwork.”199 Second, she would not be allowed to say on her website that she would not create a website for same-sex couples due to her religious convictions. That is, the wedding web designs for same-sex couples involved her speech and her silence, both wrongly compelled by government.  Each website designed by Ms. Smith would doubtless be an original, customized creation produced with dedication and specialized skill. But does this really make it her speech? As Tobias Barrington Wolf has argued, customers typically don’t pay for the privilege of facilitating a vendor’s own personal speech.200  One response is to say that it doesn’t matter whose speech it was. Current First Amendment doctrine does not require that speech be traced to a particular speaker or author to be covered or protected.  Caselaw often protects speech even when nobody would attribute the speech to the party invoking the First Amendment.  Billboard companies, for example, convey messages and have limited free speech rights201 even though nobody reasonably thinks the billboard companies endorse the specific ads or other messages that they agree to advertise. Broadcast companies have limited free speech rights,202 even though  young users’ compulsive and extended use.” States v. Meta, Case 4:23-CV-05448 (N.D. Cal, Oct. 24, 2023) complaint available at: https://ag.ny.gov/sites/default/files/court-filings/meta-multistate-complaint.pdf.  198 Solove & Hartzog, supra note 53. 199 600 U.S. 588. The Court here was quoting from the parties’ stipulations and asserting that they “lead the way” for this important analytical step.  200 Amicus Brief filed in 303 Creative, available at: ssrn:https://ssrn.com/abstract=4328872. 201 See, e.g., Metromedia, Inc. v. City of San Diego (1981) (holding that city ban on construction of billboards advertising business or its products was unconstitutional where other noncommercial signs were allowed) City Council of Los Angeles v. Tax Payers for Vincent (upholding limits on posting of political campaign advertisements on public property). 202 See, e.g., Federal Communications Commission v. League of Women Voters of California, 468 U.S. 364 (1984) (striking down rule against editorializing by public broadcasting). Electronic copy available at: https://ssrn.com/abstract=474217938  FROM GODS TO GOOGLE [28-Feb-24     38 their viewers may not attribute to these companies the content of the programs they broadcast. Private libraries can assert freedom of expression rights if the government attempts to censor their collections.203 Yet nobody reasonably thinks the library personnel or owners wrote or specifically endorsed all of the messages in Catcher in the Rye, The Color Purple, Lolita, or Rubyfruit Jungle.  Identifiable authorship of a particular message by a party invoking free speech coverage is not always required for a party to have First Amendment speech rights. On the other hand, authorial attribution has definitely mattered to the Court in prior cases.  When addressing entities that carry the speech of others, such as newspapers, libraries, and parades, the Court has assessed how closely such carriers controlled and identified with the content produced and how effectively they could dissociate themselves from unwanted messaging.204  The prospect of adequate dissociation was essential to the result in Rumsfeld v. Forum for Academic & Institutional Rights (F.A.I.R.),205 where the Court chose not to extend First Amendment protections to an unwilling mouthpiece for a government message.206 The Court held that law schools were not impermissibly forced to carry a government message or subjected to compelled silence when they were required by the Solomon Amendment to accommodate a visit by a military recruiter and provide assistance in the form of emails, notices on school billboards, and flyers –all of which involved “pure” speech—as a condition of government funding.207 This government-imposed constraint on the law school’s freedom of expressive association was seen as merely incidental to the government’s regulation of conduct.208 The Court specifically noted that the law schools had multiple other ways to convey to students their sincere disagreement with the military’s anti-gay “Don’t Ask, Don’t Tell” policy.209 They could effectively dissociate  203 Government censorship of books in public libraries too may trigger constitutional objections, based on the public’s right to receive ideas, and the authors’ free speech rights. See Board of Education, Island Trees Union Free School Dict. v. Pico, 457 U.S. 853 (1972)(plurality)(holding that although local school boards have broad control over management of local affairs, school library collection policies must comport with the First Amendment and protect the “right to receive information”). 204 See id.  205 547 U.S. 47 (2006). 206 Id.. See text accompanying notes 213-17, 236-37 infra. 207 Id. at 60-61. 208 Id. at ___ 209 Id. at ___. Justice Thomas stated in a case involving a foreign corporation’s objections to speech-related conditions on federal funding that “the mere conditioning of funds on ‘the affirmation of a belief’ tied to the purpose of a government program involves ‘no compulsion at all.’…Such a condition is ‘the reasonable price of admission to a limited government-spending program that each organization remains free to accept or reject.” Agency for International Development v. Alliance for Open Society, 140 S. Ct. 2082, ___(2020)(Thomas, J., concurring)(quoting Agency for International Development v. Alliance for Open Society, 570 U.S. 205, 226 (2013)(Scalia, J., joined by Thomas, J., dissenting). That is, government need not be viewpoint neutral; the recipient of funds can just turn down the money; and conditions on the money are not unconstitutional even if the compelled idea is attributed to the funded party. Although this is not how the majority has treated such conditions, it shows that some conservative justices view speech-sensitive conditions on funding as categorically distinct from direct regulation of private parties, as this relates to compelled speech and expressive association.  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        39 39themselves from the government message to prevent the appearance of endorsement, while allowing access to the military recruiters on equal terms with other employment recruiters. 210 This was true even though an educational program is expressive, and schools convey values through their teachings and other activities.  Comparing Rumsfeld to Becerra reveals that the Court has not been consistent in extending First Amendment rights to those who are not the primary or actual composers of a message with which they wish not to be associated, but it is nonetheless fair to say that the Court has previously viewed opportunities for dissociation to be outcome determinative.   In 303 Creative the Court was uninterested in the ample opportunities that Lorie Smith had to avoid attribution. Nothing about the Colorado law required her to add her name to a same-sex couple’s wedding website, indicate that she is the web designer, or otherwise associate her professional identity with the same-sex couples’ messages. She could have simply left her voice out, while otherwise serving the couple on equal terms to others who might hire her for their wedding messages.211 She might have simply allowed the wedding couple to speak while staying respectfully silent herself, in the way Justice Scalia once suggested should suffice for schoolchildren who might not join in a prayer at a graduation ceremony.212  How would a web wedding site reasonably be attributed to Lorie Smith, if she simply left her name off of the final product? If it would not be, why would she be treated as the relevant speaker, versus a behind-the-scenes set designer? Smith’s services would surely facilitate the speech of the couple and effectively broadcast their wedding message.  But reasonable observers would not conclude that her participation reflects a personal endorsement of their message, nor would such attribution be plausible for the hotel ballroom renter, wedding license issuer, wedding cake maker, limo driver, or napkin designer – aside from the question of which of those activities counts as speech.  Services provided behind the scenes, where the circumstances provide no reason to fear that the service provider will be publicly associated with a message objectionable to her, present a different set of concerns than a novelist being conscripted to entwine her professional and public voice with the views of her clients, or a portrait artist being required to sign her name to a glorifying portrait of Hitler.  Rather than asking the attribution question with which it had been preoccupied in Rumsfeld v. F.A.I.R, the Court relied on Barnette and Hurley v. Irish-American Gay, Lesbian, and Bisexual Group of Boston to conclude that CADA inflicted an unconstitutional coercion of the web designer’s own speech.213 But in neither of those cases did the  210 See also Pruneyard Shopping Center v. Robins, 447 U.S. 74 (1980) (holding state could require private shopping mall to provide access to speakers). 211 See Blake E. Reid, On Stipulations and Expressiveness, July 4, 2023, available at https://blakereid.org/on-stipulations-and-expressiveness/(noting that “if you don’t want to put your imprimatur on your client’s message, you could just not put your name on it!” and that “[n]one of this adds up to a concession from Colorado that the message of the hypothetical websites at issue here is actually attributable to the designer in a First-Amendment-cognizable way). 212 See Lee v. Weisman, 505 U.S. 577, 631, 637 (Scalia, J., dissenting) (arguing that to assert that students who must merely sit in respectful silence during a prayer at a public school ceremony have “somehow joined –or would somehow be perceived as having joined in the prayers is nothing short of ludicrous”).  213 515 U.S. 557 (1995).  Electronic copy available at: https://ssrn.com/abstract=474217940  FROM GODS TO GOOGLE [28-Feb-24     40 complainants have the opportunities for dissociation available to Smith.  In Hurley, the Court rejected the enforcement of Boston’s anti-discrimination law against private citizens organizing a St. Patrick’s Day parade, holding that the parade organizers could not be required to include groups that sought to impart a pro-gay message objectionable to the parade organizers.  When we consider that the private organizers initiated the parade on their own rather than at the behest of a paying customer, selected the units that would comprise the parade, and obtained a permit issued to them alone, we can appreciate the Court’s conclusion that the parade was “signed” by them. The potential attribution of the compelled message to the parade organizers was more direct than the risk of affiliation that wedding vendors might face.  Unlike the parade organizers in Hurley,214 Smith could choose to step aside, out of view, and completely obscure her participation in a message she regards as apostasy.215 And unlike the Barnette schoolchildren, she was not required to echo any vows, salute, stand, or otherwise join them.  Nor was she required to rewrite her own text or say anything at all to the public and other potential clients that might implicate her in their message. Like the law schools in F.A.I.R., she also was free to express her sincere disagreement with the government regulation in numerous other ways. To make her distance crystal clear, she could have added a disclaimer to her website designs to the effect of “I am Lorie Smith, a devout Christian website designer, and I do not approve of this same-sex marriage message.” She could readily dissociate from any endorsement of any website’s purported message. Attribution is sufficiently forceful in free speech law that we can even see it at work in Barnette, where the compelled speech took the form of a mandatory salute to the American flag, with a palm-down salute, while uttering the Pledge of Allegiance as part of an exercise meant to convey patriotic loyalty. The coercion involved conscription of both student bodies and voices, standing at attention in homage to the American flag, while intoning aloud and in unison a prescribed government message that specifically announced a first-person perspective.  But what if greater physical and other distance had been placed between the schoolchildren and the message?  Had the same children been required on a mandatory test to “State verbatim the words of the Pledge of Allegiance and explain why Americans regard it as a sacred message” the outcome would likely be different. The responses that students provide to such an exam question would not be seen as conscripting their inner consciences, religious beliefs, or political points of view. The First Amendment sin of compulsion lies not merely in the mandatory nature of a recitation but also its attributive quality.  Answering an exam question is different than standing up to salute the flag in a first-person display because it does not impute any internalized sentiment to an individual; it does not falsely suggest an inner state that is anathema to the speaker.  Similarly, the reason we do not worry about pervasive government-compelled messaging in a wide variety of places – signs about safety hazards, EEO regulations, emergency instructions, and more – is because these postings have no such attributive or internalized affirmation implications. These are so commonplace that reasonable customers understand that they are not the speech of the businesspeople with whom they transact, but rather disclosures required by the government. For Lorie Smith,  214 Id. 215 We recognize that the theological question of complicity is a much more complex one than can be fleshed out here. See, e.g., Cathleen Kaveny, Complicity With Evil, CRITERION (2003).  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        41 41though, the Court focused on compulsion as though this factor alone determined the constitutionality of the mandatory messaging allegedly entailed in the nondiscrimination measure.216  By focusing exclusively on the speaker’s behind-the-scenes “compulsion,” and ignoring the many ways in which Smith could avoid attribution of her customers’ same-sex marriage messaging, the Court unquestionably opened the door for other expressive service providers to seek exemptions from generally applicable conduct regulations where they may burden speech.  But it also injected further confusion and inconsistency into a set of principles that is particularly crucial for technology law.  The question of authorial attribution is pervasive in digital spaces.  It matters acutely when the speech interests of the service provider or platform run counter to the speech interests of the customer or user—for example, when Facebook wants to take down a post that a user wishes to keep up. Would compelling Facebook to keep up user speech be compelled speech in the sense of Barnette?217 Carefully drawn distinctions between different speakers’ rights are essential for evaluating the constitutionality of technology policy.  Take the law of the internet platform. Early discussions of platform liability asked a version of this question of authorial attribution.  Scholars and policymakers pondered when user speech should be legally attributable to a digital platform: when should a platform be liable for a user’s speech?  These early cases toggled between comparing platforms to publishers/editors like newspapers, which are legally responsible for others’ content; or to more passive distributors like bookstores, which are legally responsible for content only when they have actual knowledge of defamatory material.218 (This is more generally known as the law of intermediary liability, also referred to as a subset of vicarious liability.) Congress responded by enacting Section 230 of the Communications Decency Act, which courts have since found immunizes platforms from liability for user content in most contexts, including when they choose to take down content.219 As many have observed, the digital speech environment is one of multiple potential speakers and many competing interests.220 Platform interests and user interests in particular do not always align. Sometimes they do, and platforms will fight for user interests and rights, speech or otherwise. Indeed, in nearly every First Amendment challenge raised by technology firms, they have alleged both customers’ speech rights as well as the firm’s.221 But other times, a platform may not want to carry a particular user’s speech. A platform also may be more likely to bow to censorial pressure from  216 600 U.S. 580, ___.. 217 As Jack Balkin has recently put it, on the internet free speech is a triangle. Jack Balkin, Free Speech Is a Triangle, 118 COLUM. L. REV. 2011 (2018). 218 Compare Cubby, Inc. v. CompuServe Inc., 776 F. Supp. 135 (S.D.N.Y. 1991) (finding that CompuServe was not liable as a publisher of the content posted online by its users) with Stratton Oakmont, Inc. v. Prodigy Services Co., 23 Media L. Rep. (BNA) 1794 (N.Y. Sup. Ct. 1995) (treating internet service provider like a publisher that is liable for the content that it publishes because it had made editorial decisions). 219 See, e.g., JEFF KOSSEFF, THE TWENTY-SIX WORDS THAT CREATED THE INTERNET (2019).  220 See, e.g., Balkin, Free Speech is a Triangle, supra note 216.  221  Rozenshtein, Silicon Valley’s Speech, supra note 14, at 4 (collecting sources).  Electronic copy available at: https://ssrn.com/abstract=474217942  FROM GODS TO GOOGLE [28-Feb-24     42 government entities where a user might instead push back.222 These tensions between platform interests and user interests have led states like Texas and Florida, motivated by fears (justified or not) about the unfair treatment of politically conservative speech by internet platforms, to attempt to treat internet platforms more like telecommunications companies, enacting laws attempting to impose “must-carry” and transparency requirements on certain companies.223 In cases in both the Fifth Circuit and Eleventh Circuit, courts have evaluated the significance of authorial attribution in assessing First Amendment arguments by platforms against these types of regulation.  The Eleventh Circuit found that social-media platforms have a First Amendment right to moderate content that had been violated by Florida’s law. 224  It found that existing caselaw permitting the government to compel private actors to “host” others’ speech was inapplicable for two reasons: (1) because Florida’s must-carry law interfered with social media companies’ “own message” (as opposed to the users’ messages it was hosting), 225 and (2) because social-media platforms’ content moderation is inherently expressive, unlike, say, law-school recruiting services.226 That is, it mattered to the court both that the content-moderation choices of social-media companies were more speechlike than the actions of other non-digital hosts in earlier cases (law schools, shopping malls), and it also mattered that the must-carry requirements interfered with the platforms’ speech in particular, as opposed to the users’ speech. The Eleventh Circuit considered but rejected the State’s proposal that “in order to trigger First Amendment scrutiny a regulation must create a risk that viewers or listeners might confuse a user’s and the platform’s speech.”227 In other words, it found that social-media companies as hosts had sufficiently distinct messages conveyed through their content moderation as to have their own distinct speech rights. It was speech, and it was theirs. By contrast, for the Fifth Circuit, the relevant speech belonged to the users, not the platforms, and this made all the difference. The Fifth Circuit found that while “the State may not force a private speaker to speak someone’s else message… the State can regulate conduct in a way that requires private entities to host, transmit, or otherwise facilitate speech.”228 The court of appeals contrasted Miami Herald, where a newspaper could not be compelled to convey messages it opposed,229 with Prune Yard, where a private shopping mall could not refuse to serve as a forum for the speech of others because (1) few would think that the pamphleteers’ speech was coming from the mall itself, because it was open to the public; and (2) the mall could readily disavow any  222 See Derek E. Bambauer, Against Jawboning, 100 MINN. L. REV. 51, 87 (2015) (describing the many pressures that governments exert on private actors without directly regulating their speech). 223 See Reid, Uncommon Carriage, supra note 173.  224 NetChoice v. Attorney General of Florida, No. 21-12355 (11th Cir. 2022) at 31. 225 Id. at 34-35 (“a law that requires the platform to disseminate speech with which it disagrees interferes with its own message and thereby implicates its First Amendment rights”). 226 Id. at 35-36. See also id. at 35 (“unlike law-school recruiting services, [social media companies] are in the business of disseminating curated collections of speech”) 227 Id. at 39. 228 NetChoice v. Paxton (5th Cir.) at 20 (“Were it otherwise, no government could impose nondiscrimination requirements on, say, telephone companies or shipping services.”). 229 Id. at 21. Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        43 43affiliation with the speakers’ messages.230 The Fifth Circuit also specifically noted that in Hurley, the parade case, it mattered to the Supreme Court that the parade organizers were “intimately connected” to the parade’s message—it was speech, and it was theirs.231 By contrast, in F.A.IR., hosting military recruiters was neither speech nor the law school’s.232 The Fifth Circuit concluded that “[u]nder these precedents, a speech host must make one of two showings to mount a First Amendment challenge. It must show that the challenged law either (a) compels the host to speak or (b) restricts the host’s own speech. The Platforms cannot make either showing.”233 Instead, the court found that the platforms were mere conduits, and had repeatedly argued as much in other contexts.234 To be clear, we are not advocating here for a particular outcome as to authorial attribution in the platform law cases; our goal is only to explain that the developments in 303 Creative are highly relevant for platform regulation. 303 Creative puts an unnecessarily heavy thumb on the scale by assuming that once something is deemed to be “pure speech” involving words and images that it doesn’t matter whose speech it actually is—even when the consumer/speaker has interests that run counter to the speech host.235 As we see from the appellate divide in the NetChoice cases, whose speech it is—the platforms or their users—can be outcome determinative.  C. Finding that Government Regulation Conscripts the Conscience   Closely related to the attribution and dissociation issue is a question about the degree to which a speaker’s expressive autonomy is burdened. The majority in 303 Creative not only treated CADA’s application to the web designer as an infringement of her speech, bypassing more persuasive ways of viewing that expression.  It also viewed it as a particularly egregious infringement – a complete preemption of her expressive autonomy on a matter of conscience and religious belief. The majority offered no discussion of the avenues available to Ms. Smith to both comply with the law and express her authentic views. As F.A.I.R. shows, however, when determining whether a regulation imposes undue coercion of expression, it should matter whether a speaker has  230 Prune Yard, 447 U.S. at 87-88. NetChoice v. Paxton (5th Cir.) at 22-23. According to the Fifth Circuit, a later Supreme Court case, PG&E, further distinguished Prune Yard by noting that the mall’s owner was not itself trying to speak, and that the access requirement in Prune Yard was content-neutral. Id. at 24. See also PG&E plurality, 475 U.S. at 21 (Marshall, J.). 231 Id. at 26. Hurley:, 515 U.S. 557, 576 (“[W]hen dissemination of a view contrary to one’s own is forced upon a speaker intimately connected with the communication advanced, the speaker’s right to autonomy over the message is compromised“); see also id. at 573 (emphasizing that “a speaker has the autonomy to choose the content of his own message,” including by “decid[ing] what not to say”) (quotation omitted). See also Netchoice 5th Circuit at “The Court instead carefully limited its holding to a speech host (like a parade organizer or composer) who is “intimately connected” with the hosted speech” (at 30). 232 Id. at 27. 233 Id. 234 Id. at 28-29. “Unlike newspapers, the Platforms exercise virtually no editorial control or judgment. The Platforms use algorithms to screen out certain obscene and spam-related content. And then virtually everything else is just posted to the Platform with zero editorial control or judgment.” 235 303 Creative, 600 U.S. 580.   Electronic copy available at: https://ssrn.com/abstract=474217944  FROM GODS TO GOOGLE [28-Feb-24     44 other expressive options to distance themselves and their true beliefs from the message expressed. Smith had several, as we already have noted. She could disavow the speech to make clear she did not endorse the message.  She could have posted a message announcing: “I don’t want to create websites for same-sex weddings but Colorado is making me.” 236  This would avoid the attribution problem discussed above and would also reduce Smith’s compliance burden for the very pragmatic reason that potential customers might choose to go elsewhere.237 If Smith is allowed to post a notice making it quite unlikely that she would ever have to make a website for a same-sex wedding, her compliance with CADA becomes even less conscriptive.238  These expressive alternatives put Smith in a different position than the parade organizers in Hurley.  Once the Boston officials deemed a parade to be a public accommodation, the organizers had no alternative to design a parade without a pro-gay message. Smith did not face this same problem – she remained free to create an infinite number of additional websites that proclaim her competing view.  If conscription is the core of what mattered in Barnette—that the government impermissibly forced schoolchildren to publicly state something they did not believe—then the degree to which compelled disclosures impinge on that concern should matter. To be sure, the Court has been inconsistent in its willingness to consider the degree of expressive autonomy left open to a speaker bringing a free speech challenge to a law with which they do not wish to comply.   Even modest burdens on religious actors have been deemed impermissible in recent cases.239 For example, it was of no import in  236 We recognize that she would have to comply with the provision in CADA that prohibits “directly or indirectly” speaking in such a way that makes a customer feel “unwelcome, objectionable, unacceptable, or undesirable.”  600 U.S. at ___ . At oral argument counsel for the state conceded that a website designer “can put anything it wants on a standardized website, even if that includes a denunciation of same-sex marriage.” Transcript p. 67. And also that one could post "Made With Love by Amber, who believes that a valid marriage is a union between one man and one woman.”  237 As Justice Alito forecasted, “that website designer is not going to be serving a same-sex couple if the website designer puts that on the website.” Transcript p. 68.  238 A striking moment in the oral argument reveals the extent to which Smith’s free speech theory obliterates the concept of authorial attribution.  Upon questioning from Justice Kagan about where we would find the vendor’s message given that she could provide a website to “Mike and Mary,” and then “Mike and Mark” come along and say, I love that website you did, we’ll take that exact design just change the names and the date, counsel responded that “the announcement of the wedding itself is a concept that she believes to be false.” Transcript pp. 38-40.  239 For example, in Little Sisters of the Poor Saints Peter and Paul Home v. Pennsylvania, 591 U.S. ___(2020), the Court respected the “complicity with evil” burden alleged by the Catholic Order in holding that the government had the authority to promulgate rules that provided religious and moral exemptions to that Affordable Care Act. Although the case was decided on statutory grounds, it reflects the Court’s deference to religious actors’ claims about burdens with little scrutiny. See also Zubick v. Burwell, 578 U.S. __(2017)(arguing that merely submitting a notice to obtain accommodation notice to obtain an exemption from compliance with the Affordable Care Act was a substantial burden under the Religious Freedom Restoration Act). In the COVID cases, the Court emphasized the severity of the burdens imposed on religious people when public health laws prohibited indoor gatherings of stated numbers of persons that included religious gatherings for worship. See note 142 supra.  In Establishment Clause cases, in contrast, the burdens imposed on nonobservant persons when government Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        45 45Kennedy v. Bremerton—which held that a high school football coach could not be prevented from praying on the field immediately after the game—that the coach could pray after the event, in his car, on his drive home, during his lunch break, or anywhere else beyond the school and his official duties.240 Nor was the Court in Becerra moved by the fact that posting the required notices, no larger than letter-sized copy paper, left the pregnancy centers free to denounce abortion and cajole their pregnant patients into childbirth in myriad ways.   The Court’s readiness to find speech compulsion unduly coercive in religious expression cases like Becerra and 303 Creative will have serious implications for technology law. As we have outlined, technology policy often relies heavily on mandatory disclosures–whether as future-proofing, as a form of softer-touch regulation, as accountability necessary to collaborative governance, or as a check on power imbalances. The Court has greatly increased the possibility that mandatory disclosure requirements as well as any regulation that seems to compel speech will be found unconstitutional.241 This is a problem for privacy regulation, efforts to reform content moderation, and, increasingly, regulatory efforts aimed at artificial intelligence. As we have noted, the entire privacy law regime in the United States for a long time relied on “notice and choice”: disclosing information about privacy practices to individuals so they could ostensibly choose whether or not to partake in a product or service.242 The Fair Information Practice Principles (FIPPs) on which all data privacy laws since the 1970s have been based also centrally depend on disclosure as an element of fair process–that is, notice and an opportunity to be heard, of sorts, for data subjects.243 But what stops a Court like the one that decided Becerra and 303 Creative from seeing any of these requirements as compelled speech conscripting the beliefs of website designers and other programmers?  This same concern arises in debates over artificial intelligence. Not only do most proposed regulations of AI involve some kind of transparency requirement, but this may require companies to revise their algorithms in order to make them more transparent. There is debate among AI researchers about explainable AI, and the different ways of creating algorithms that are scrutable to outsiders.244 If regulators require the purveyors of these algorithms to change their code to be more transparent,  adopts policies or practices that promote or endorse religion are given much less weight. For example, in cases in which the Court has held that public funds not only can go to private religious schools, but that such schools have a free exercise right not to be denied these funds on equal terms with secular schools, the Court gives no attention to the ways in which nonobservant taxpayers too may feel complicit with “evil” insofar as some of these schools may teach students ideas and practices that are deeply offensive and hurtful, including religious-inspired views about reproductive rights, gender equality, race and ethnicity, other religions, or LGBTQ rights. Relatedly, the third-party harms of mandatory accommodation of religious conduct or speech are relatively underweighted or ignored entirely.  240 597 U.S. ___(2022). 241 See text accompanying notes 40-57 supra. 242 See e.g. Solove, supra note 23; Waldman, supra note 26. 243 Margot E. Kaminski & Jennifer M. Urban, The Right to Contest AI, 121 COLUM. L. REV. 1957 (2021)  (summarizing the history of FIPPs).. 244 For an overview, see Andrew Selbst & Solon Barocas, The Intuitive Appeal of Explainable Machines, 87 FORDHAM L. REV. 3 (2018); see also A. Schubbach, Judging Machines: Philosophical Aspects of Deep Learning, 198 SYNTHESE 2 (2018).  Electronic copy available at: https://ssrn.com/abstract=474217946  FROM GODS TO GOOGLE [28-Feb-24     46 we can imagine those purveyors appealing to this Court’s willingness to find compelled speech at every turn. A similar concern arises with proposals to regulate cybersecurity, content moderation, cryptocurrency, and intellectual property—any law that might require general transparency.245 The 303 Creative Court’s willingness to find compelled speech where it need not is a dangerous omen for the coming wave of technology regulations where the lightest-touch intervention would be something like a transparency mandate.   D. From Strict Scrutiny to Categorical Invalidity  Even if one accepts that commercial web design is pure speech and not a regulable form of expressive conduct; that the speech in question is attributable to the web designer and not to the customer; and that requiring such service to be provided on equal terms is unduly coercive compelled speech; there should be no full stop at the end of this sequence. Up until 303 Creative, it was very clear that the next doctrinal step would be the application of strict scrutiny, assessing whether the state can provide a compelling interest for the restriction and show that the regulation is narrowly tailored to that interest.   The most stunning and doctrinally disruptive move in 303 Creative was its utter failure to apply strict scrutiny. To get there, the Court not only had to leap-frog O’Brien and ignore Giboney’s admonition that speech may be regulated when it is integral to illegal conduct. It had to rewrite the bedrock free speech principle that even when government regulates protected speech in a content-specific way, it still may prevail if it has a compelling reason for doing so. Even though strict scrutiny is “the most demanding test known to constitutional law,” 246 it is a rule that nonetheless provides for serious review.247 Yet the Court simply waved away compelling and place-sensitive interests by asserting the non-answer that “this is compelled speech.” It invoked the lessons of Barnette, Hurley, and Dale not only as relevant to the component questions addressed above, but also as a substitute for the application of strict scrutiny. It did this sub silentio, without expressly instructing when or why certain instances of compelled speech may be subject not to strict scrutiny but to a categorical rule of invalidity.  Barnette predated the Court’s tiers-of-review approach to constitutional law, so the formal absence of steps that we now recognize as strict scrutiny is unsurprising. In any  245 See Jennifer Urban on 512; others on 230. E.g. evelyn douek. 246 See, e.g., City of Boerne v. Flores, 521 U.S. 507, 534 (1997)(discussing rigors of strict scrutiny).     247 See Williams-Yulee v. Fla. Bar, 575 U.S. 433 (2015); Holder v. Humanitarian Law Project, 561 U.S. 1 (2010); Burson v. Freeman, 504 U.S. 191 (1992); Austin v. Mich. Chamber of Commerce, 494 U.S. 652 (1990); Buckley v. Valeo, 424 U.S. 1 (1976). Although the universe of cases in which the Court has deemed strict scrutiny to be met is small, it proves the point. Moreover, if the Roberts Court continues to apply freedom of speech to zones where regulation serves compelling government interests, and in which the primary justifications for robust free speech are attenuated –that is, that lie well beyond the iconic streetcorner speaker engaged in political discourse, or the schoolchildren conscripted in voice and body to salute the flag –then this small set of cases can and should grow. It should not be allowed to simply dodge scrutiny as it did in 303 Creative or create a newly-minted right to “no scrutiny” in alleged compelled speech cases. Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        47 47event, however, none of these cases eschewed strict scrutiny in any explicit sense. Even Judge Tymkovich, who dissented from the appellate court’s majority opinion in 303 Creative, did not read those cases to bypass strict scrutiny. Rather, he asserted that the Colorado law should be subject to strict scrutiny as both speech compulsion and a content-based restriction on speech. His proposed strict scrutiny analysis was granular: he accepted that the state has a compelling interest in eliminating discrimination and ensuring equal access to publicly available goods and services, but argued that Colorado did not have a compelling interest in ensuring access to a particular person’s services, no matter how unique they might be.  Justice Gorsuch, in contrast, conducted no such analysis. He first embraced the appellate court’s conclusion that CADA regulates “pure speech” and that the speech in question was Smith’s. He then proceeded to “part ways with the Tenth Circuit only when it comes to the legal conclusions that follow. While that court thought Colorado could compel speech from Ms. Smith consistent with the Constitution, our First Amendment precedents laid out … teach otherwise.”248 He trotted out a parade of horribles that might follow if the government were allowed to “compel anyone who speaks for pay on a given topic to accept all commissions on that same topic—no matter the underlying message—if the topic somehow implicates a customer's statutorily protected trait.”249 He nodded briefly toward “the vital role public accommodations laws play in realizing the civil rights of all Americans,” but then emphasized the ways in which contemporary public accommodations laws are broader than the historical versions that were recognized as serving a compelling interest and historically imposed on “common carriers and places of traditional public accommodation like hotels and restaurants.”250 He added that “some States… have expanded the reach of these nondiscrimination rules to cover virtually every place of business engaged in any sales to the public. …[and]… to prohibit more forms of discrimination….[A]pproximately half the States have … expressly prohibit discrimination on the basis of sexual orientation. And… this is entirely “un-exceptional.”251 Several things are notable about these passages. First, recall that the Court had already characterized Colorado’s objective in strenuously pejorative terms as the desire to “excis[e] certain ideas or viewpoints from the public dialogue” and “to coercively eliminate dissenting ideas” (our emphases).252 Especially against that backdrop, this passage does little to acknowledge that there are people whose interests in being treated with dignity and equality in the marketplace are advanced by the expanded public accommodations laws. Nor does the discussion assess how those interests weigh against the interests of religious speakers like Smith: precisely the difficult balancing that strict scrutiny demands.  Second, this passage emphasizes the expansion of public accommodations laws beyond the versions that were originally described as advancing a “compelling interest.” If the majority wants to suggest that the state interest in prohibiting discrimination on the basis of sexual orientation is weaker than the state interest in prohibiting racial  248 303 Creative, 600 U.S. at 588 (invoking Hurley, Dale, and Barnette).  249 Id. 250 Id. 251 Id. at ___(citations omitted)..  252 Id. at 588. Electronic copy available at: https://ssrn.com/abstract=474217948  FROM GODS TO GOOGLE [28-Feb-24     48 discrimination, it should go ahead and say so. Or if it wants to convey that states have compelling interests only with regards to the classic public accommodations of yore, the hotels and the restaurants and the riverboats and the trains,253 then it should be candid about this significant limitation.  Instead, this passage conveys these doctrinal amendments subtly, without confronting directly whether Colorado has a compelling interest in prohibiting discrimination on the basis of sexual orientation across the spectrum of businesses defined by the state as public accommodations.  The passage concludes with statements that are utterly nonresponsive to a core issue: “no public accommodations law is immune from the demands of the Constitution,”254 and “public accommodations laws can sweep too broadly when deployed to compel speech.”255 Both statements are obviously true, but do not illuminate what the constitutional demands were in this situation and whether these were met.  Specifically, did Colorado have a compelling interest in requiring wedding website designers to provide their services on equal terms to same sex and opposite sex couples alike? Was CADA sufficiently narrowly tailored?  This complete elision of strict scrutiny analysis makes 303 Creative an outlier even among the other recent cases that protect religious expression in doctrinally troubling ways. Becerra, as discussed above, laid the groundwork for the view that any compliance burden is tantamount to undue coercion – but the Court there at least applied strict scrutiny, preserving the role of that framework in assessing burdens on constitutional rights and keeping open the possibility that other burdens of this nature might satisfy the demanding standard. It did not apply the sort of automatic invalidity that we see in 303 Creative.256  The same was true in Kennedy, where the Court protected a high school football coach’s right to pray on the field.257 Treating Kennedy as a case in which the Court was merely following its own doctrinal rules was a stretch—it chastised the lower courts for following Supreme Court precedent that had never been overruled.258 But again, the Court at least worked within the rubric of strict scrutiny,259 evaluating and then rejecting the school’s proffered interests.260  By refusing to employ strict scrutiny at all,  253 Transcript p. 61 254 600 U.S. at ___. 255 Id. 256 Nor did Fulton, as explained above. See text accompanying notes 143-45 supra. 257 597 U.S. ___(2022). 258 Andrew Koppelman, Religious Liberty As A Judicial Autoimmune Disorder: The Supreme Court Repudiates Its Own Authority in Kennedy v. Bremerton School District, 74 Hastings L.J. 1751, 1753 (2023) (describing Kennedy as a case illustrating that the “Court is so predisposed to find discrimination against religion that it declared it to be present in a case where the discriminator was obeying the Court's own commands.”).  259 Although with the slightest note of equivocation. Justice Gorsuch wrote: “Under the Free Exercise Clause, a government entity normally must satisfy at least ‘strict scrutiny,’ showing that its restrictions on the plaintiff ’s protected rights serve a compelling interest and are narrowly tailored to that end. A similar standard generally obtains under the Free Speech Clause.” Kennedy, 142 S. Ct. 2407, 2426 (2022). If strict scrutiny, described in City of Boerne v. Flores as “the most demanding standard known to constitutional law” is the least that a government would have to satisfy, what would it be at most? A possibility is that Justice Gorsuch is laying the groundwork for categorical invalidity rules of the sort seen in 303 Creative.  260 It must be said, however, that to dismiss the school’s concern about undue influence the majority had to characterize the activity as a “private prayer,” a manipulation of the facts directly Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        49 49303 Creative pushes third-party harms even further out of the frame. 261 The Court was unwilling to even consider the state interests as it had done –albeit nominally262 -- in prior cases that were profoundly solicitous of religious expression and that modified First Amendment free exercise and free speech precedent in various ways.263 Instead, a public accommodations law was deemed unenforceable against a covered business without serious discussion of the state’s interest in nondiscrimination mandates. Although the application of strict scrutiny may not have saved CADA, it would have forced the Court into a more honest, complete, and respectful assessment of what Colorado was trying to do. The Court has for years been vigorously policing equality of treatment between religious and non-religious people, with much greater concern about the former than the latter. But in 303 Creative, the Court collapsed almost to a vanishing point the weight that equality concerns are given on the other side of the ledger. Had the Court grappled with Colorado’s interest in CADA in the way that strict scrutiny requires, it would have needed to address whether and when equality is a sufficiently compelling interest to justify the speech burdens that CADA imposes.264 Instead, it suggests that the  contradicted by the photographic evidence provided by the dissent. For more on the “deceitful narrative spun by counsel” attempting to portray that the coach “was disciplined for holding silent, private prayers,” see Kennedy v. Bremerton Sch. Dist., 4 F.4th 910, 912 (9th Cir. 2021). Judge Smith then provided a chart that refutes each “unmoored claim” offered by the coach with “what the record actually shows.” Id. at 917-920.  261 Republican Party of Minnesota v. White, 536 U.S. 765 (2002).. 262 We say “nominally” for several reasons. These earlier cases described what was a once-compelling government reason not to accommodate or support religious speech and conduct -- the avoidance of an Establishment Clause violation --as unconstitutional discrimination. Government efforts to avoid direction of its resources to private parties that discriminate against protected class members may be redescribed as government hostility. See, e.g., Fulton, 593 U.S.___ (2021).  263 Kennedy “disregards how vindicating a school official's religious claim will redound to the detriment of students who are religious minorities at a vulnerable time in their lives.”  Justin Driver, Three Hail Marys: Carson, Kennedy, and the Fractured détente over Religion and Education, 136 HARV. L. REV. 208, 255 (2022).  Kennedy completed a long project of undermining the ability of government actors to invoke Establishment Clause-concerns as a compelling reason to prevent government financial or other support of religion. Government efforts to avoid entanglement or endorsement no longer offer a persuasive justification to treat religion differently from other ideas or conduct; rather, the failure to support religious actors or institutions on equal terms with others is treated as government hostility to religion, versus a reasonably neutral way of avoiding sectarian uses of its resources in ways that could undermine other shared values, including tolerance. 264 As Yoshino explains, once we understand the government interest to include protecting against the dignitary harms of humiliation and stigma rather than simply ensuring equal access to the material goods and services, we can see that laws such as CADA could not be any more narrowly tailored, and the alternative vendor argument loses force. See Yoshino, note 136 supra. That the individuals denied service may simply “go elsewhere,” though, should be no conversation-stopping answer to the dignity and other harms this conduct inflicts. After all, in Katzenbach v. McClung, Black patrons could buy barbeque from Ollie’s Barbeque through the back door; they just were denied service in the main dining area. 379 U.S. 294 (1964). See Xiao Wang, Religion as Disobedience 376 VAND. L. REV. 999 (2023) (discussing how doctrine has greatly eased religious plaintiffs’ burden of demonstrating that a government rule infringes their sincere Electronic copy available at: https://ssrn.com/abstract=474217950  FROM GODS TO GOOGLE [28-Feb-24     50 government no longer is permitted to demand compliance with public accommodations laws whenever they may compel protected expression, no matter how compelling the government’s asserted equality interests might be.  This bypass of strict scrutiny is plainly threatening to technology law because it eliminates judicial inquiry into whether the regulation in question furthers a compelling government interest due to grave potential harms to others. The central policy question in many upcoming technology battles will not be whether there are some speech interests in question—as we have shown, there often will be—but rather how those speech interests should be weighed against compelling government interests aimed at protecting people online.  Consider a few examples. There is an ongoing debate about law enforcement access to encrypted data.265 In the wake of the San Bernardino terrorist attack, the FBI asked Apple to write software that would give the firm access to the suspect’s communications.266 Apple argued that any requirement to de-encrypt its devices or services would violate the First Amendment (on behalf of the firm and its users).267 The case was ultimately dropped because the FBI paid a private firm to hack into the phone, but this is unlikely to be the last time that government concerns for national security conflict with speech rights, either of a technology firm or its clients. The resolution of this so-called “going dark” debate—as with any tension between speech rights and national security—is impossible to imagine without a full assessment and weighing of government regulatory interests. This built-in kind of cost-benefit analysis has long been one of the strengths of First Amendment jurisprudence.  We need this kind of careful balancing in the age of AI more than ever. Regulators increasingly demand that purveyors of large AI systems offer transparency into how they operate.268 In the U.S., those transparency demands may face First Amendment challenges, where one of the crucial questions will be—or rather, ought to be—whether the government can sufficiently justify its requirement. This kind of balancing is not only the right way to navigate thorny speech issues; it is the right way to develop technology policy.  And what about generative AI? Suppose that someone creates a botnet that generates speech designed to sway an election. As the head of the government’s chief cybersecurity office recently wrote, “Generative AI in the hands of adversaries could threaten each part of the electoral process, including the registration of voters, the  religious practices, and imposing stricter demands on government to justify these infringements, leading to lawbreaking without repercussions). Cf. Stephanie H. Barclay, First Amendment “Harms,” 95 Ind. L. J. 331 (2020) (offering a way based on the Coase Theorem that she argues would better account for third party harms caused by accommodation of religious actors).  265 See Kristin Finklea, Encryption and the “Going Dark” Debate, Congressional Research Service (July 20, 2016). 266 See Alan Z. Rozenshtein, Surveillance Intermediaries, 70 STAN. L. REV. 99, 102-3 (2018). 267  See In re Search of an Apple iPhone Seized During the Execution of a Search Warrant on a Black Lexus IS300, No. ED 15-0451M, 2016 WL 618401, at *1 (C.D. Cal. Feb. 16, 2016). 268 As an example, transparency is a bedrock principle of the EU’s AI Act See EU AI Act, Key Issues: Transparency Obligations (“A key priority of the EU AI Act is transparency due to its ability to enable citizens to understand AI systems’ design and use, as well as enable accountability for decisions made by companies and public authorities.”) available at: https://www.euaiact.com/key-issue/5.  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        51 51casting of votes, and the reporting of results.”269 It seems hard to imagine state actions aimed at generative election interference that do not raise at least some speech concerns; we are talking about political speech, after all. Yet the government ought to be able to take sensible steps to protect elections. What counts as “sensible” in this context ought to depend on the government’s ability to convince a court that its interests are compelling and its actions well-tailored.   303 Creative portends a future in which we will have to fight these inevitable battles without the disciplining force of strict scrutiny.     CONCLUSION  At the dawn of the internet, one could tell a happy story about how American free speech culture and American democracy were both synergetic with this new technology. If the First Amendment was in significant part premised on the idea that there is too little speech in the world, then a technology to make it easier for people to speak was a First Amendment technology.270 One judge put it in exactly these terms in an early case: “As the most participatory form of mass speech yet developed, the Internet deserves the highest protection from governmental intrusion.”271 That case, ACLU v. Reno, would eventually make its way to the Supreme Court, posing the Court’s first test of an internet regulation.272 In a move indicative of things to come, the Court relied on the First Amendment to strike down aspects of the revised Communications Decency Act.273 In short, the First Amendment was a friend to the early internet.  Today, however, the story one might tell about democracy, free speech, and technology is somewhat bleaker. In the wake of the 2016 Russian disinformation campaign, our Covid misinformation problems, “the Big Lie,” and more—the internet appears more antagonistic to democracy. And where before the First Amendment was the guardian of both a healthy democracy and a healthy internet, today the First Amendment looks like the biggest obstacle to just about every effort to regulate even simple digital harms. It is against this backdrop that we raise the alarm about the Court’s clumsy and heady-handed religious speaker cases, epitomized by 303 Creative. In this light, the move from gods to Google is not just a problem for speech law or technology law; it is a problem for democracy. It is not just that the Court seems intent on creating a First Amendment that would undermine national consensus on basic technology regulation, though that is true. But we also worry the Court is crafting, intentionally or not, a First Amendment that could frustrate the well-functioning of the  269 Jen Easterly, Scott Schwab, Cait Conley, Artificial Intelligence’s Threat to Democracy, FOREIGN AFFAIRS (Jan. 3, 2024) https://www.foreignaffairs.com/united-states/artificial-intelligences-threat-democracy.  270 Jack Goldsmith argues that one of the two pillars of the U.S.’s early internet policy was an anti-censorship principle, which the U.S. sought to export: “American-style freedom of speech and expression on the global internet.” Jack Goldsmith, The Failure of Internet Freedom, Knight First Amendment Institute Essay (June 13, 2018) https://knightcolumbia.org/content/failure-internet-freedom.  271 Am. C.L. Union v. Reno, 929 F. Supp. 824, 883 (E.D. Pa. 1996).  272 521 U.S. 844 (1997). 273 Mark S. Kende, The Supreme Court’s Approach to the First Amendment In Cyberspace: Free Speech as technology’s Hand-Maiden, 14 CONST. COMMENTARY 465, 475 (1997). Electronic copy available at: https://ssrn.com/abstract=474217952  FROM GODS TO GOOGLE [28-Feb-24     52 republic—precisely the opposite of what Madison and Brandeis and so many others believed would be the case.274 Those early free speech sages were not wrong about the fit between free speech and democracy; but it turned out that the fit is contingent on certain social conditions and calls for great care by a subtle and context-sensitive Court.275  We can imagine several distinct kinds of democratic costs of moving from gods to Google. The first, perhaps most obvious, set of concerns turns on the nature of information as a key democratic input. If digital speech spreads falsehoods, for example, and this leads to the electorate being uninformed, this undermines the democratic process.276 Another possibility is that online speech fora inhibit sensible debate; even if the information is all true and valid, people may be reluctant to debate important ideas for fear of being shamed, doxxed, or worse. In this scenario, the online environment harms democracy by encouraging conformity and discouraging the deliberation that the Founders thought essential to democracy. These are serious concerns about the online environment degrading expression in a way that threatens democracy. If the First Amendment sits by, or even encourages these developments, the First Amendment looks more like a threat to democracy than a bulwark.  But the Court’s religious speaker cases threaten democracy in other ways too. Privacy is an essential component of democracy. If the online environment erodes personal privacy, democracy suffers. We have explained multiple ways that the Court’s religious speaker cases threaten privacy laws. If any of the blame for the degradation of  274 Whitney v. California, 274 U.S. 357, 375 (1927) (Brandeis, J., concurring) overruled in part by Brandenburg v. Ohio, 395 U.S. 444 (1969) (“freedom to think as you will and to speak as you think are means indispensable to the discovery and spread of political truth”). See also James Madison, the drafter of the First Amendment, told the House: “If we advert to the nature of republican government, we shall find that the censorial power is in the people over the government, and not in the government over the people.” 4 Annals of Cong. 934 (1794). See also Ashutosh Bhagwat, The Democratic First Amendment, 110 NW. L. REV. 1097 (Developing the thesis that “the primary—albeit not necessarily the only—reason why the First Amendment protects freedom of speech is to advance democratic self-governance”);  CASS R. SUNSTEIN, DEMOCRACY AND THE PROBLEM OF FREE SPEECH 121–65 (1993)(contending that free speech is a “precondition” for democracy); Ashutosh Bhagwat, Details: Specific Facts and the First Amendment, 86 S. CAL. L. REV. 1, 33-35 (2012) (summarizing cases). Cf.  Toni M. Massaro & Helen Norton, Free Speech and Democracy, a Primer for the 21st Century Reformers, 54 CAL-DAVIS L. REV.  1631 (2021)(expressing concern about free speech doctrine that threatens to undermine democracy, rather than reinforce it and discussing means by which to prevent this corrosion). 275 Among these is the idea that the greatest threat to speech comes from the state, not private actors. As Jack Balkin notes, recent decades have seen “significant changes in the practices and technologies of free expression, changes that concern a revolution in the infrastructure of free expression. That infrastructure, largely held in private hands, is the central battleground over free speech in the digital era.” Jack Balkin, Old-School/New-School Speech Regulation, 127 HARV. L. REV. 2296, 2296 (2014). See also Andrew Keane Woods, Private Platforms, Public Law,_107 MINN. L. REV. 1249, 1260 (2023)(“Not only does the First Amendment not guarantee the freedom of political speech on a platform like Facebook, it does not even articulate reasonable standards that might apply to Facebook, like the requirement that the firm make its speech decisions transparent, fair, and sensible.”) 276 See Stephan Lewandowsky, Ullrich K.H. Ecker, John Cook, Sander van der Linden, Jon Roozenbeek, Naomi Oreskes, Misinformation and the Epistemic Integrity of Democracy, 54 CURR. OPINION IN PSYCH. (Dec. 2023).  Electronic copy available at: https://ssrn.com/abstract=474217928-Feb-24] FROM GODS TO GOOGLE        53 53digital privacy belongs to the First Amendment, then the rosy story about the synergy between free speech and self-government suddenly seems less convincing. The same can be said for other online harms that erode democracy, like surveillance and discrimination.  And then there are the online tools that threaten the longstanding organizational backbone of our democracy. As Rick Pildes has argued, “American democracy faces profound challenges in our era,” thanks to a communications revolution that has created intense political polarization.277 The “fragmenting effects” of today’s digital tools are complex, but can be ascribed to a few trends.278 One principal development is that individual voters have the media reach and the organizational tools to create micro-parties within parties, effectively fracturing political consensus within a single party.279 Then there are the effects on politicians, who are also capable of wielding these tools. There are a growing number of free-agent politicians--individual politicians who can appeal to a national audience and fundraise without the help of the party, making them hard to constrain and corral.280 Pildes sees these developments as destabilizing to democracy. Once again, the Court seems intent on creating a First Amendment that will stand in the way of fixing this problem lying at the very heart of the democratic process.  Whatever one thinks of these various harms, one thing is clear: navigating the triad of novel technologies, the First Amendment, and the democratic process are the most challenging issues we face as a society. Addressing these challenges will require the kind of careful, contextual reasoning that is mindful of precisely those distinctions we find wanting in the Court’s recent religious speaker cases—distinctions between speech and conduct, between customers and vendors, between regulation and conscription, and between individual rights and governmental interests. This kind of careful reasoning is necessary to maintain a sensible free speech and free exercise regime, of course, but it is also—and this has been overlooked—essential as the court’s gaze inevitably shifts from messiahs to machines, from supplicants to cyborgs, from priests to platforms, from gods to Google.     277 Richard H. Pildes, Democracies in the Age of Fragmentation, 110 CAL. L. REV. 2051 (2022). 278 Id. at 2066. 279 Id. at 2063-65. 280 Id. at 2066. Electronic copy available at: https://ssrn.com/abstract=4742179

Clean Full Text

Language

Doi

Arxiv

Mag

Acl

Pmid

Pmcid

Pub Date

Pub Year

Journal Name

Journal Volume

Journal Page

Publication Types

Tldr

Tldr Version

Generated Tldr

Search Term Used

Reference Count

Citation Count

Influential Citation Count

Last Update

Status

Aws Job

Last Checked

Modified

Created