Comments to the FTC on suppression of accuracy in AI systems

Comments of the Abundance Institute
On the proposed policy statement concerning the suppression of accuracy in artificial intelligence systems
Before the Federal Trade Commission

AI POLICY STATEMENT; MATTER NO. P264200
DOCKET NUMBER: FTC-2026-0859
JULY 2026

The Abundance Institute welcomes the opportunity to comment on the Federal Trade Commission’s proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems.¹ The Abundance Institute is a mission-driven nonprofit organization dedicated to creating a cultural and policy environment in which emerging technologies can develop, thrive, and expand widespread human prosperity. Its scholars have testified before Congress, submitted comments to federal agencies, and published extensively on artificial intelligence, competition, and the legal barriers that can slow innovation.²

Artificial intelligence is a general-purpose technology developed and delivered through national and global networks. A growing patchwork of U.S. state AI laws is imposing costs and product-design mandates; some state laws have effects far beyond the enacting state’s borders. The Abundance Institute has previously documented this extraterritorial effect and urged the federal government to use its lawful tools – including litigation, agency advocacy, and, where Congress has supplied it, preemption – to protect interstate innovation.³

While the states are piling on new laws, existing federal consumer-protection laws already apply to AI. Consumers should receive what AI providers promise. An AI company should not advertise a system as truth-seeking, accurate, objective, or responsive to the user’s goals while secretly steering it toward a different ideological objective. If this occurs, the FTC’s existing Section 5 authority can address it.

The draft statement is therefore directionally right about deception, but insufficiently clear about the limits of the Commission’s authority and incomplete in its preemption guidance. These flaws carry costs. Overbroad “steering” language invites enforcement against ordinary product design, and a conclusory preemption discussion gives courts nothing to apply, leaving the Executive Order’s goal of a workable national framework unmet. To avoid these costs, the final statement should:

• Explain the familiar deception framework and how it applies to the commercial claims and net impressions surrounding deployed AI products. Avoid implying that model development or product design, standing alone, counts as a representation to consumers.
• Explain that conflict preemption would be an as-applied challenge raised in court. Section 5 preempts only state obligations that mandate a materially deceptive communication with no available truthful way to comply with both state and federal law.
• Clarify that Section 5 prohibits materially misleading commercial conduct but does not impose a general duty to make AI systems accurate, neutral, unbiased, or maximally helpful. It does not license the government to select a preferred viewpoint or supervise the editorial judgments of private speakers.
• Commit to filing targeted amicus briefs in cases testing whether state laws conflict with Section 5.

I. BACKGROUND

President Trump issued Executive Order 14365 (“EO”) in December 2025 to advance a minimally burdensome national policy framework for AI and reduce the costs of discordant state regulation.⁴ Section 7 of the order directed the Commission to explain how FTC Act Section 5 applies to AI model development and, in particular, how state laws that require alterations to truthful model outputs may conflict with the FTC Act.⁵ The Commission responded with this proposed enforcement policy statement.

The President’s goal matters: AI systems often serve users in many states from shared models, infrastructure, datasets, and product interfaces. Fifty or more materially different rules governing the development or use of every AI model could force a provider to fragment its product, withdraw it from smaller markets, or apply the most restrictive state’s rule nationwide. And that patchwork of state law is growing fast. In 2024, state legislators introduced 635 AI-related bills and enacted 99; in 2025, they introduced more than 1,000; so far in 2026, more than 1,800.⁶ Not every state AI law overreaches, and many govern local uses or state agencies. But comprehensive state laws aimed at model development or nationally distributed systems can create substantial burdens on interstate commerce.

AI does, however, present legitimate consumer-protection concerns. Providers market AI systems as tools that solve problems, answer questions, and advance users’ objectives. The Commission’s draft statement collects examples of companies presenting their products as accurate, reliable, evidence-grounded, or truth-seeking. Against that backdrop, a provider may create a misleading net impression if it intentionally degrades factual responses to serve an undisclosed ideological objective. The specific ideology does not matter. Section 5 protects consumers from material deception, not from one side of a political debate.

These two concerns, regulatory fragmentation and undisclosed steering, overlap only in a subset of cases. A state law could, theoretically, require a provider to make a statement that is materially deceptive under federal law. Other state laws regulate documentation, testing, risk management, disclosures, or a particular use of AI. Such laws might be ill-considered but are unlikely to require deception. And sometimes a state law may pressure a provider’s choices without compelling any consumer-facing message. The final statement should draw these distinctions.

II. THE DRAFT STATEMENT CORRECTLY APPLIES SETTLED DECEPTION PRINCIPLES, BUT THE FINAL STATEMENT SHOULD TIE LIABILITY TO COMMERCIAL CLAIMS AND DEPLOYMENT

A. SECTION 5 REACHES MATERIALLY MISLEADING CLAIMS AND PRACTICES INVOLVING AI

The draft statement correctly begins with the Commission’s established deception test. An act or practice is deceptive when a material representation, omission, or practice is likely to mislead consumers acting reasonably under the circumstances.⁷ The Commission evaluates the net impression of the communication, not isolated words. Express claims are not required: implied claims and half-truths can mislead. And an objective claim may require substantiation proportionate to what reasonable consumers understand the advertiser to promise.

Those principles apply to AI just as they apply to other technologies. If a provider advertises a model as “truth-seeking,” claims a measured accuracy rate, promises that an agent will follow the user’s instructions, or implies that an AI service will perform a professional task at a stated level, the provider must have a reasonable basis for the claim and must not undermine it through undisclosed practices. A conspicuous qualification may change the net impression. Fine print that contradicts the headline ordinarily will not.

The Commission’s central example fits comfortably within this framework. Suppose a provider repeatedly represents that its system will give the most accurate answer available within its technical and resource constraints. The provider then secretly instructs the deployed system to replace a disfavored fact with a politically preferred falsehood. If the undisclosed instruction is likely to matter to a reasonable consumer’s decision to purchase, use, or rely on the service, the resulting mismatch between promise and product can be deceptive. Whether the motive is profit, politics, public or private pressure, or state-law compliance, the provider must still represent its product’s features and capabilities truthfully.

B. THE FINAL STATEMENT SHOULD FOCUS ON INTENTIONAL AND COVERT STEERING

The draft statement needs qualification, however, where it asserts that “[T]he Commission believes that AI companies that steer the outputs of their AI systems toward unexpected objectives, and away from the objectives set by or reasonably expected by users, are likely to deceive consumers in violation of Section 5 of the FTC Act.”⁸ That sentence (and the paragraph that follows) characterizes all steering as likely deceptive. But as defined in the draft, “steering” covers a wide range of potential behavior, much of it beneficial rather than deceptive.

INTENTIONAL

The final statement should clarify that it is primarily concerned with intentional steering of AI outputs rather than incidental or accidental steering.

Intent is not an element of deception under Section 5. A provider is responsible for the net impression its claims convey to reasonable consumers, whatever it meant to say. But Commission precedent has long treated intent as powerful evidence that the elements are met.⁹

Most relevant here, the Commission’s cases distinguish between products or services intentionally designed to differ from their advertising and products that simply vary in normal use. The first category involves a deliberate, material gap built into the product itself. AT&T sold “unlimited” data plans while a secret program throttled customers past a set threshold.¹⁰ Volkswagen sold “clean diesel” cars built to behave one way in emissions tests and another on the road.¹¹ In cases like these, every sale involves a deceptive failure to disclose the gap between the promise and product. If advertising claims lead reasonable consumers to choose an AI system expecting politically neutral results, and the provider intentionally manipulates those results, that too could be a deceptive failure to disclose.

The Commission treats ordinary product variance differently. Many products and services vary in quality or other dimensions through unintended errors. In these cases, the Commission asks whether the company overpromised consistency; the violation, if any, is the false claim of consistency, not the inconsistency itself.

As the draft already notes, AI hallucinations fall into this latter category. They are not deceptive because they “stem not from a design decision to prioritize objectives contrary to users’ reasonable expectations, but from the technological and resource limitations AI systems necessarily reflect.”¹² That logic extends beyond hallucinations to other AI system imperfections. Unless paired with materially misleading claims of reliability, best-efforts performance that falls short of perfection is not deceptive.

The final statement should therefore clarify that its core concern is intentional manipulation of AI system results.

COVERT

The draft should also focus on covert behavior. Hidden steering techniques such as system and meta prompts and chain-of-thought management are easy for AI system developers to deploy yet hard for users to detect. If such techniques contradict a company’s promises to its users, they could be deceptive.

By contrast, overt steering mechanisms range from educational to irritating, but they are rarely deceptive. Refusals are a good example. A refusal occurs when an AI service declines to answer a user’s question or perform an assigned task, usually with an explanation.¹³ Refusals are an important mechanism that AI companies use to manage different kinds of risk, from copyright to legal compliance to safety.¹⁴ When poorly calibrated, refusals can create a bad experience for users. But even then, they are unlikely to be deceptive. Such interactions are usually “easily evaluate[d]” by the consumer because the refusal is immediate and visible.¹⁵ The individual prompts involved are also “frequently purchased,” and are generally “inexpensive”: dozens or hundreds of prompts per day, usually for pennies apiece.¹⁶ All of this means “market incentives place strong constraints” on how AI system providers use refusals.¹⁷

The Commission should therefore replace categorical language suggesting that all unexpected steering is “likely to deceive” with a more conventional formulation: unexpected steering may be deceptive when it causes a material mismatch between a provider’s practices and the net impression reasonably conveyed to consumers, and this mismatch is most likely for steering practices that are intentional and covert. That wording aims the FTC’s deception enforcement at the most problematic behavior.

III. THE FINAL STATEMENT SHOULD EXPLAIN MORE PRECISELY WHEN SECTION 5 MAY PREEMPT A STATE REQUIREMENT

A. THE DRAFT STATEMENT STATES A CONCLUSION BUT DOES NOT SUPPLY A USABLE TEST

Under the EO, the final statement “must explain the circumstances under which State laws that require alterations to the truthful outputs of AI models are preempted by the Federal Trade Commission Act’s prohibition on engaging in deceptive acts or practices affecting commerce.”¹⁸

The current draft does not.

The draft explains that, while the FTC Act contains no express preemption clause, state law is impliedly preempted to the extent it conflicts with a federal regulatory scheme. That is correct: under the Supremacy Clause, valid federal law prevails when simultaneous compliance is impossible or when state law stands as an obstacle to federal purposes.¹⁹ But saying that a state law requiring deception “obviously conflicts” with Section 5 does not “explain the circumstances” in a way that helps courts, states, or regulated parties determine when a state law actually requires deception.

A policy statement cannot bind the courts that will ultimately determine any preemptive effect. But a thorough, consistent, and persuasive explanation of the federal scheme and the practical conflict could shape how those courts rule.²⁰

B. THE COMMISSION SHOULD OUTLINE A PARTICULARIZED, AS-APPLIED CONFLICT FRAMEWORK

The final statement should explain that conflict preemption by Section 5 requires four elements:

• A state obligation. Identify the statutory or regulatory text, the covered actor, the relevant enforcement interpretation, and the conduct required or forbidden.
• That affects a consumer-facing communication. Identify the representation, omission, ranking, label, suppression, or presentation that compliance would produce in a particular commercial product or transaction.
• So that the communication violates Section 5. Explain the reasonable consumer interpretation, why it would be misleading, and why the difference is material under settled deception doctrine.
• And there is no truthful compliance path. Explain why attribution, qualification, disclosure, an alternative product configuration, or another lawful means cannot preserve a truthful net impression while satisfying state law.

This approach resembles the Supreme Court’s caution in antitrust preemption cases. A state statute is not displaced merely because it permits conduct that might be unlawful under federal law; the conflict must arise from what the law mandates, authorizes, or irresistibly pressures private parties to do in the relevant application.²¹ The analogy is imperfect but useful. That a regulated firm might choose a deceptive response to a state law does not mean the law requires deception.

The Colorado example in the draft shows why this matters. The draft reasons that a duty to prevent algorithmic discrimination may incentivize a developer to suppress accurate outputs.²² That possibility deserves scrutiny, but predicting an incentive does not prove that a particular state command compels a materially deceptive consumer-facing claim or leaves no truthful route to compliance. Before declaring a conflict, courts will likely require defendants claiming preemption to establish all four elements.

As a result, facial preemption of state laws by Section 5 will likely be rare. A court is unlikely to invalidate a state law in all applications unless the law necessarily produces materially deceptive conduct across all or substantially all of them. A final statement that reflects this modesty serves the Executive Order better than a too-ambitious argument for broad preemption. A narrow preemption theory that wins as applied builds durable precedent; a broad theory that courts reject leaves the state patchwork intact and the federal framework weaker than before.

C. PREEMPTION SHOULD APPLY ONLY WHEN NO TRUTHFUL MEANS OF COMPLIANCE EXISTS

The statement should explain that a state law is not preempted simply because some ways to comply would violate Section 5. If a provider can comply with both, there is no conflict. A genuine conflict could arise if a state law requires a provider to state a falsehood under its own name, forbids the provider from adding information necessary to avoid a misleading half-truth, or compels a presentation that necessarily creates a materially false net impression. In that situation, Section 5 forbids what state law requires.

By contrast, there is ordinarily no conflict when a provider can:

• identify the content as a state-required notice rather than the provider’s own factual claim;
• separate the mandated content from the provider’s answer or recommendation;
• add clear and conspicuous truthful context; or
• offer a compliant product configuration without contradicting its commercial claims.

The draft already recognizes that disclosures can alter consumer expectations. The same principle applies in preemption analysis. If an adequate disclosure about the state legal requirement cures the alleged deception, simultaneous compliance remains possible and conflict preemption is unwarranted. If the state forbids an adequate cure or requires the provider to adopt the misleading statement as its own, preemption is more plausible.

D. THE POLICY STATEMENT SHOULD CLARIFY HOW SECTION 5 PREEMPTION OPERATES

The final statement should spell out how Section 5 preemption would operate. First, it should explain that the legal source of any preemptive effect is the FTC Act as applied to an actual conflict, not the policy statement itself. Second, the statement should note that the Commission may bring a Section 5 case against a covered company whose compliance with state law results in a materially deceptive practice. And third, the statement should highlight that a regulated party may argue in litigation that a particular state requirement is preempted because it compels conduct federal law forbids.

E. THE COMMISSION SHOULD DEVELOP THE DOCTRINE THROUGH TARGETED AMICUS PARTICIPATION

Courts will flesh out Section 5’s preemptive effect on state AI laws, likely in cases state attorneys general bring against private companies. The Commission should participate in such cases, and the statement should commit to doing so. When private litigants raise a preemption argument, the Commission should file an amicus brief explaining the preemptive effect of Section 5 and when it applies. The Commission has used amicus participation to address the preemptive scope of other consumer-protection regimes.²³

Amicus participation is especially suitable here. The Commission’s expertise applied to varied fact patterns would help courts distinguish genuine compulsion from speculative pressure, identify disclosures that avoid deception, and calibrate relief. A developed judicial record will also help the Commission update the policy statement over time.

IV. THE FINAL STATEMENT MUST BE CLEAR ABOUT WHAT SECTION 5 DOES NOT AND CANNOT DO

A. SECTION 5 DOES NOT IMPOSE A GENERAL MANDATE OF ACCURACY, NEUTRALITY, OR OBJECTIVITY

Section 5 is not a federal truth code. It does not require an AI model to be accurate in every answer, guarantee the absence of hallucinations, maximize a single conception of usefulness, or satisfy a government-defined standard of political neutrality. The draft statement appropriately recognizes that ordinary errors caused by technological or resource limitations are not themselves Section 5 violations, although misrepresentations about their frequency or likelihood may be deceptive.²⁴ The final statement should elevate that point from a footnote to the main text.

Nor can a “neutrality” requirement replace the deception elements. Every useful AI product embodies choices about sources, objectives, refusal rules, uncertainty, style, and tradeoffs. A provider may build a system that is cautious or permissive, partisan or nonpartisan, religious or secular, family-oriented or adult, provided its commercial representations do not materially mislead. Consumers may reward, reject, compare, criticize, or compete with those choices. Section 5 polices the honesty of the offer. It does not dictate the offer.

This limitation applies even when the Commission finds a provider’s choices objectionable. A hidden ideological objective is only relevant because it may contradict a material express or implied claim. Ideology is not itself the violation. The same analysis should apply to undisclosed steering from the left, right, center, or any other political perspective.

The final statement should also avoid treating “accuracy” as self-defining in domains where reasonable people dispute values, predictions, classifications, or the framing of a question. Enforcement should focus on verifiable claims and demonstrably misleading practices. When the alleged falsity depends on the Commission selecting among contested political or moral views, the agency is no longer applying ordinary deception law.

B. SECTION 5 ENFORCEMENT MUST RESPECT THE FIRST AMENDMENT

False or misleading commercial speech does not receive the same protection as truthful, nonmisleading speech, and the Commission may enforce disclosure and substantiation duties within its statutory authority.²⁵ But the First Amendment still marks the boundary of that authority.

The Supreme Court has repeatedly held that government may not compel a private speaker to convey the government’s preferred message. It has also recognized that selecting, organizing, prioritizing, and presenting third-party speech can involve protected editorial judgment, and that government may not impose what officials consider the correct balance of private expression.²⁶ Those principles constrain any policy statement about suppressing, replacing, ranking, labeling, or otherwise altering AI output.

The Commission can remain on firm ground by anchoring enforcement to a commercial transaction and a materially misleading net impression. The statement should expressly explain that Section 5 cannot be used to:

• require a provider to generate lawful content it chooses not to provide;
• require political balance or viewpoint neutrality;
• treat a disclosed value judgment or editorial policy as a false factual claim;
• regulate noncommercial research, open-source publication, or private model development merely because of the ideas a model may express; or
• pressure a provider to adopt the Commission’s preferred safety, alignment, or ideological rules.

Compelled disclosures also require care. Even in commercial settings, a mandatory disclosure should be factual, uncontroversial, justified by a real risk of deception, and no more burdensome than reasonably necessary.²⁷ A requirement that every AI response carry a persistent government-scripted warning about contested ideological concepts could itself raise serious constitutional concerns. The final statement should recommend disclosures as one possible way for a provider to avoid a misleading net impression rather than announce a mandatory universal script.

C. THE FINAL STATEMENT SHOULD PRESERVE ROOM FOR LAWFUL PRODUCT DESIGN AND USER CHOICE

The Commission should distinguish deception from legitimate multi-objective design. Providers may prioritize safety, privacy, legality, reliability, age-appropriateness, cybersecurity, or other product goals. They may offer different models with different values or capabilities. They may refuse prompts, express uncertainty, attach sources, or invite users to choose among models. Those design decisions can improve accuracy and other qualities consumers value.

The Commission should use deception enforcement to promote transparency and competition, not impose a federal design mandate. Providers should describe material product characteristics honestly. Independent evaluators, users, and rivals can then test those claims and offer alternatives. Open and competitive AI markets make it easier for consumers to select systems aligned with their preferences and harder for any single provider’s hidden objective to dominate the information environment.

V. RECOMMENDED CLARIFYING LANGUAGE

The final statement could capture these principles with language along the following lines:

Section 5 does not require an artificial intelligence system to be accurate, neutral, unbiased, or responsive to every lawful request. It prohibits covered persons from making materially misleading representations, omissions, or practices in or affecting commerce. Intentional and covert steering of AI results can be deceptive under Section 5 when it contradicts a material commercial claim. Product design, model development, training, fine-tuning, and editorial judgment are relevant to deception only insofar as a covered person’s commercial communication or practice creates a materially misleading net impression for reasonable consumers about such practices.

A state requirement may conflict with Section 5 when, as applied to a covered commercial actor, it requires a representation, omission, or practice that is materially deceptive under settled Section 5 principles and leaves no reasonably available truthful and nonmisleading means of complying with both state and federal law. Any preemptive effect arises from the federal statute, is determined by a court on a concrete record, and extends no further than the actual conflict.

This statement does not occupy the field of artificial-intelligence regulation; displace state requirements merely because they concern AI; regulate noncommercial speech; compel viewpoint neutrality; or authorize the Commission to prescribe the content of lawful AI outputs. The Commission will evaluate claims and practices in context, consistent with the First Amendment and the limits of its statutory authority.

This language would carry out the Executive Order’s request while preserving the settled boundaries of consumer-protection law.

VI. CONCLUSION

The Commission should finalize a carefully revised statement that treats intentional and covert steering as deceptive only when it contradicts a material commercial claim. It should pair that rule with a particularized, as-applied preemption test requiring proof that no truthful compliance path remains; recognition that courts determine any preemptive effect on a concrete record; and an express disclaimer of any federal accuracy or neutrality mandate. Those limits would protect consumers without making the FTC a national editor of artificial intelligence.

Respectfully submitted,

Neil Chilson
Head of AI Policy, Abundance Institute

---

Footnotes

1. Federal Trade Commission, Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems, 91 Fed. Reg. 41638 (July 7, 2026) (“Draft Statement”), https://www.govinfo.gov/content/pkg/FR-2026-07-07/pdf/2026-13628.pdf. ↩
2. Abundance Institute, About (last visited July 23, 2026), https://abundance.institute/about. ↩
3. Neil Chilson, Abundance Institute, Clearing the Path for AI: Federal Tools to Address State Overreach (Sept. 2025), https://abundance.institute/ our-work/clearing-the-path-for-ai. ↩
4. Exec. Order No. 14365, Ensuring a National Policy Framework for Artificial Intelligence, 90 Fed. Reg. 58499 (Dec. 11, 2025), https://www.govinfo.gov/content/pkg/FR-2025-12-16/pdf/2025-23092.pdf. ↩
5. Id. at 58500, § 7. ↩
6. Chilson, supra note 3, at section titled “The Patchwork Problem: How State AI Laws Threaten Economic Growth and Interstate Commerce.” See also Multistate.ai, Artificial Intelligence (AI) Legislation Tracker 2026: All 50 States (last visited July 28, 2026), https://www.multistate.ai/artificial-intelligence-ai-legislation. ↩
7. Federal Trade Commission, Policy Statement on Deception, 103 F.T.C. 174, 175, 182-83 (1984), appended to In re Cliffdale Associates, Inc., 103 F.T.C. 110 (1984) (“Deception Policy Statement”), https://www.ftc.gov/legal-library/browse/ftc-policy-statement-deception. ↩
8. Draft Statement, 91 Fed. Reg. at 4164. ↩
9. Thompson Medical Co., 104 F.T.C. 648 (1984) (intentional efficacy claims for Aspercreme mean those claims are material); Kraft, Inc. v. FTC, 970 F.2d 311 (7th Cir. 1992) (“[I]ntent to convey a claim is one of three categories qualifying for a presumption of materiality”). ↩
10. Federal Trade Commission, Press Release, FTC Says AT&T Has Misled Millions of Consumers with ‘Unlimited’ Data Promises (Oct. 28, 2014), https://www.ftc.gov/news-events/news/press-releases/2014/10/ftc-says-att-has-misled-millions-consumers-unlimited-data-promises. ↩
11. Federal Trade Commission, Press Release, FTC Charges Volkswagen Deceived Consumers with Its “Clean Diesel” Campaign (Mar. 29, 2016), https://www.ftc.gov/news-events/news/press-releases/2016/03/ftc-charges-volkswagen-deceived-consumers-its-clean-diesel-campaign. ↩
12. Draft Statement, 91 Fed. Reg. at 41641 n.46. ↩
13. Alexander von Recum et al., Cannot or Should Not? Automatic Analysis of Refusal Composition in IFT/RLHF Datasets and Refusal Behavior of Black-Box LLMs (2024), https://doi.org/10.48550/arXiv.2412.16974. ↩
14. Neil Chilson, How to Prevent Woke AI – Part Two at heading Inference-Time Filters and Moderation Layers (Post-Generation Screens) (May 29, 2025), https://outofcontrol.substack.com/i/164714345/7-inference-time-filters-and-moderation-layers-post-generation-screens. ↩
15. Deception Policy Statement, supra note 7, at 5. ↩
16. Id. ↩
17. Id. ↩
18. Exec. Order No. 14,365, supra note 4, 90 Fed. Reg. at 58,500, § 7. ↩
19. U.S. Const. art. VI, cl. 2; Murphy v. National Collegiate Athletic Association, 584 U.S. 453, 477 (2018); Arizona v. United States, 567 U.S. 387, 399 (2012). ↩
20. Wyeth v. Levine, 555 U.S. 555, 576-77 (2009) (explaining that an agency has “no special authority to pronounce on preemption absent congressional delegation,” while recognizing that the weight of an agency’s explanation of how a state law impacts a federal scheme depends on the explanation’s thoroughness, consistency, and persuasiveness). ↩
21. Rice v. Norman Williams Co., 458 U.S. 654, 659, 661 (1982) (requiring an “irreconcilable conflict” and examining whether state law mandates, authorizes, or places irresistible pressure on private parties to violate federal antitrust law). ↩
22. Draft Statement, 91 Fed. Reg. at 41641 & nn.42-44. ↩
23. See, e.g., Federal Trade Commission, FTC Files Brief in Jones v. Google in Support of Appeals Court Ruling that COPPA Does Not Preempt Plaintiffs’ State Privacy Claims (May 19, 2023), https://www.ftc.gov/news-events/news/press-releases/2023/05/ftc-files-brief-jones-v-google-support-appeals-court-ruling-coppa-does-not-preempt-plaintiffs-state. ↩
24. Draft Statement, 91 Fed. Reg. at 41641 n.46. ↩
25. Central Hudson Gas & Electric Corp. v. Public Service Commission, 447 U.S. 557, 563-64 (1980). ↩
26. Moody v. NetChoice, LLC, 603 U.S. 707, 731-33, 740-42 (2024); 303 Creative LLC v. Elenis, 600 U.S. 570, 586-89, 596 (2023). ↩
27. National Institute of Family and Life Advocates v. Becerra, 585 U.S. 755, 768-78 (2018); Zauderer v. Office of Disciplinary Counsel, 471 U.S. 626, 651 (1985) ↩