FEDERAL TRADE COMMISSION’S PROPOSED POLICY STATEMENT CONCERNING THE

Agency decision

Ask Donna

What actually matters in this document.

Text

SUPPRESSION OF ACCURACY IN ARTIFICIAL INTELLIGENCE SYSTEMS

July 1, 2026

Summary

Artificial Intelligence (AI) is reshaping how Americans consume information, educate our

children, and perform our jobs. 1 As they have marketed their remarkable breakthroughs to the

public, AI companies have spent years representing explicitly and implicitly that their systems aim

to produce the best output—output that faithfully and accurately achieves users’ stated objectives

and the built-in objectives that users expect in the AI system—that is possible within their

technological and resource constraints. Because of these representations and the inherent nature of

the products and services in question, consumers have a reasonable expectation that AI systems

aim to give truthful and accurate outputs. Consumers have no basis to believe that AI systems aim

to produce outputs that are distorted by undisclosed ideological objectives.

Nonetheless, an AI company might be tempted to alter or steer the output of its systems

contrary to consumers’ reasonable expectations for various reasons, including attempted

compliance with a state law, such as Colorado’s recently revised Artificial Intelligence Act. But

steering an AI system in this manner may deceive consumers in violation of Section 5 of the FTC

Act. That is true even if the deceptive steering is done in an effort to comply with state laws. Of

course, a company may be able to avert potential deception by making truthful, non-misleading

representations about the aims of its model. But such representations would need to make clear

that the AI company is prioritizing objectives different than those consumers requested or would

otherwise expect.

II.

Background

Artificial intelligence is an umbrella term covering a universe of different tools and systems

that are now used across nearly every sector of the economy and in most people’s daily lives. 2

Regardless of whether one has in mind modern large language models, AI applications applying

large-language models to particular purposes, or a not-yet-developed superintelligence, the

hallmark of a successful AI system is an ability to accurately “make predictions, recommendations,

Executive Office of the President, Winning the Race: America’s AI Action Plan at 1 (July 2025),

https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf [hereinafter “AI Action

Plan”]; The Council of Economic Advisors, Executive Office of the President, Artificial Intelligence and the Great

Divergence at 1 (Jan. 2026), https://www.whitehouse.gov/wp-content/uploads/2026/01/Artificial-Intelligence-andthe-Great-Divergence-5.pdf.

See, e.g., Exec. Order No. 14319, Preventing Woke AI in the Federal Government, at § 1, 90 Fed. Reg. 35389,

35389 (July 23, 2025) (“Artificial intelligence . . . will play a critical role in how Americans of all ages learn new

skills, consume information, and navigate their daily lives.”).

or decisions” for its user consistent with a given objective. 3 Their utility can then be judged by

how well the solution matches consumers’ objectives.

On a fundamental level, these products and services solve problems for people. For many

Americans, they are becoming part and parcel of daily life, relied upon for research, analysis, and

advice on both personal and business issues. This development reflects consumers’ confidence that

AI companies are bound by the same rules as other American companies: they will deal with

consumers in good faith and with no hidden agenda, and they will not work in the shadows to

punish those who hold opinions contrary to theirs. In large part, that confidence itself reflects the

basic fact that the United States is the global standard-bearer for AI technologies. 4 That American

companies dominate every layer of the AI ecosystem should not be taken for granted. Geopolitical

rivals are investing heavily in this sphere, hoping to inject their own companies and values into

the marketplace. American dominance is thus about more than winning some abstract race. It is

about ensuring that Americans can continue to feel that their values are being respected as they

interact with, and benefit from, an AI-powered economy.

Under the President’s leadership, the Trump-Vance Administration has taken decisive steps

to sustain America’s global AI dominance 5 by removing regulatory barriers that impede AI

innovation. 6 Excessive AI regulation would undermine American AI supremacy by deterring and

suppressing the same ingenuity responsible for making American AI great. 7 At the same time,

however, as President Trump’s National Policy Framework for Artificial Intelligence recognizes,

responsible AI innovation can co-exist with prudent guardrails. For example, AI technologies

should protect children and empower parents. 8 And most relevant to this statement, AI

technologies should not be used to silence or censor lawful expression or dissent. 9 Importantly,

President Trump’s proposed approach is a national AI framework, protecting innovation and

competition by providing national regulatory clarity and certainty and avoiding a balkanized or

15 U.S.C. § 9401(3).

See J.D. Vance, Remarks by the Vice President at the Artificial Intelligence Action Summit in Paris, France (Feb.

11, 2025), www.presidency.ucsb.edu/documents/remarks-the-vice-president-the-artificial-intelligence-actionsummit-paris-france [hereinafter “Vance AI Summit Remarks”].

See, e.g., AI Action Plan, supra note 1; Exec. Order No. 14318, Accelerating Federal Permitting of Data Center

Infrastructure, 90 Fed. Reg. 35385, 35385 (July 23, 2025) (prioritizing the rapid and efficient buildout of AI data

centers and the infrastructure that power them by easing federal regulatory burdens); Exec. Order No. 14319, 90

Fed. Reg. at 35389 (noting agencies have an “obligation not to procure [AI] models that sacrifice truthfulness and

accuracy to ideological agendas”); Exec. Order No. 14320, Promoting the Export of the American AI Technology

Stack, 90 Fed. Reg. 35393, 35393 (July 23, 2025) (establishing a coordinated national effort to support the American

AI industry by promoting the export of full-stack American AI technology packages).

See, e.g., Exec. Order No. 14148, Initial Rescissions of Harmful Executive Orders and Actions, 90 Fed. Reg. 8237,

8240 (Jan. 20, 2025), https://www.govinfo.gov/content/pkg/FR-2025-01-28/pdf/2025-01901.pdf (revoking Exec.

Order No. 14110, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, 88 Fed. Reg. 75191

(Oct. 30, 2023)); Exec. Order No. 14179, Removing Barriers to American Leadership in Artificial Intelligence, 90

Fed. Reg. 8741, 8741–42 (Jan. 23, 2025), www.govinfo.gov/content/pkg/FR-2025-01-31/pdf/2025-02172.pdf.

Exec. Order No. 14179, 90 Fed. Reg. at 8741; Vance AI Summit Remarks, supra note 4.

The White House, National Policy Framework for Artificial Intelligence (Mar. 20, 2026).

Id.

patchwork regulatory approach driven by the states—or, most dangerously, imposed by certain

anti-innovation state governments on the rest of the country.

America’s AI Action Plan and other critical executive actions strike that balance between

accelerating innovation and protecting Americans, 10 but anti-innovation states’ recent efforts to

regulate AI are concerning. The growing number of enacted and proposed state AI laws threatens

to create a patchwork of regulatory regimes and compliance challenges for American companies. 11

For example, some states have enacted laws regulating the outputs of various AI models, ultimately

requiring American companies to embed “ideological bias within [their AI] models.” 12

10F

11F

On December 11, 2025, President Trump signed Executive Order 14365, “Ensuring a

National Policy Framework for Artificial Intelligence.” 13 The Executive Order builds on the

Administration’s prior work to encourage adoption of AI and remove regulatory barriers, aiming

to “sustain and enhance the United States’ global AI dominance” by establishing a “minimally

burdensome national policy framework for AI—not 50 discordant State ones.” 14 The Executive

Order directs the Commission to issue this enforcement policy statement clarifying the application

of Section 5 of the Federal Trade Commission Act (“FTC Act”) to AI models, and in particular,

address how state laws requiring alterations to the accurate outputs of AI models can conflict with

the requirements of the FTC Act. 15

12F

13F

14F

III.

The Commission’s Authority under Section 5 of the FTC Act

Nearly ninety years ago, Congress gave the Commission the authority to protect consumers

from “unfair or deceptive acts or practices in or affecting commerce.” 16 Importantly, Congress did

not provide state-law safe harbors—even where a State’s laws shape the provision of those

products or services, companies must comply with Section 5.

15F

Section 5 of the FTC Act’s prohibition on deceptive acts or practices secures consumers’

right to truth in the marketplace. 17 In our 1983 Policy Statement on Deception, we said that “the

Commission will find deception if there is a representation, omission or practice that is likely to

16F

See supra note 5.

See The White House, Fact Sheet: President Donald J. Trump Ensures a National Policy Framework for Artificial

Intelligence (Dec. 11, 2025), https://www.whitehouse.gov/fact-sheets/2025/12/fact-sheet-president-donald-j-trumpensures-a-national-policy-framework-for-artificial-intelligence/ [hereinafter “AI National Policy Fact Sheet”] (“State

legislatures have introduced over 1,000 different AI bills[.]”); Exec. Order No. 14319, 90 Fed. Reg. at 35389.

See, e.g., Exec. Order No. 14365, Ensuring a National Policy Framework for Artificial Intelligence, 90 Fed. Reg.

58499, 58499 (Dec. 11, 2025) (“For example, a new Colorado law banning ‘algorithmic discrimination’ may even

force AI models to produce false results in order to avoid a ‘differential treatment or impact’ on protected groups.”);

AI National Policy Fact Sheet, supra note 11 (“States such as California and Colorado are considering requiring AI

companies to censor outputs and insert left-wing ideology in their programming.”). Colorado has since materially

revised the law referred to by this Executive Order, but the new version poses many of the same concerns. See infra

note 43.

Exec. Order No. 14365, 90 Fed. Reg. at 58499.

Id.

Id. at 58500 (Section 7).

Wheeler-Lea Act, Pub. L. No. 75-447, 52 Stat. 111 (1938) (codified in relevant part at 15 U.S.C. § 45(a)(1)).

In re Int'l Harvester Co., 104 F.T.C. 949, 1056 (1984).

mislead the consumer acting reasonably in the circumstances, to the consumer’s detriment.” 18 This

test, which “undergird[s] all deception cases,” consists of three elements. 19 “First, there must be a

representation, omission or practice that is likely to mislead the consumer.” 20 “Second, we examine

the practice from the perspective of a consumer acting reasonably in the circumstances.” 21 And

“[t]hird, the representation, omission or practice must be a material one,” such that it is “likely to

affect the consumer’s conduct or decision with regard to a product or service.” 22

17F

18F

19F

20F

21F

Federal courts have adopted the same principles in assessing whether an act or practice is

deceptive. 23 Courts have also held that the Commission’s deception authority covers implied

misrepresentations, half-truths, 24 and instances where only a “significant minority” of consumers

are misled. 25 Moreover, courts have recognized that the Commission’s deception authority

requires the entity to substantiate the representation it makes about its products or services with

sufficient evidence. 26 At its core, the Commission’s deception authority is based on “wellestablished principles of advertising law and common sense.” 27

22F

23F

24F

25F

26F

The Commission has applied the principles underlying its Section 5 deception authority to

address false or misleading claims in a multitude of factual circumstances and across a wide variety

FTC Policy Statement on Deception (“Deception Policy Statement”), 103 F.T.C. 174, 183 (1984), (appended to In

re Cliffdale Assocs., Inc., 103 F.T.C. 110 (1984)), https://www.ftc.gov/legal-library/browse/ftc-policy-statementdeception.

Id. at 175.

Id.

Id. Material information affects a consumer’s decision to purchase a product but may also affect other kinds of

consumer conduct. Id. at 182 n.45 (citing the Commission’s complaint in Volkswagen of America, 99 F.T.C. 446

(1982)).

See, e.g., FTC v. Pukke, 53 F.4th 80, 104 (4th Cir. 2022); FTC v. Moses, 913 F.3d 297, 306 (2d Cir. 2019); FTC v.

E.M.A. Nationwide, Inc., 767 F.3d 611, 631 (6th Cir. 2014).

See, e.g., Novartis Corp. v. FTC, 223 F.3d 783, 787 (D.C. Cir. 2000); Fanning v. FTC, 821 F.3d 164, 170–72 (1st

Cir. 2016); Kraft, Inc. v. FTC, 970 F.2d 311, 322 (7th Cir. 1992).

ECM Biofilms, Inc. v. FTC, 851 F.3d 599, 610 (6th Cir. 2017) (quoting POM Wonderful, LLC v. FTC, 777 F.3d

478, 490 (D.C. Cir. 2015)). A small proportion of consumers, such as 10%, can constitute a “significant minority.”

See, e.g., Telebrands Corp., 140 F.T.C. 278, 325 (2005), aff’d, 457 F.3d 354 (4th Cir. 2006) (10.5% is significant);

Firestone Tire & Rubber Co. v. FTC, 481 F.2d 246, 249 (6th Cir. 1973) (10-15% is significant).

See, e.g., POM Wonderful, 777 F.3d at 490-91; FTC v. Direct Mktg. Concepts, Inc., 624 F.3d 1, 8 (1st Cir. 2010);

FTC v. Nat’l Urological Grp., Inc., 645 F. Supp. 2d 1167, 1190 (N.D. Ga. 2008), aff’d, 356 F. App’x 358 (11th Cir.

2009).

FTC v. Com. Planet, Inc., 878 F. Supp. 2d 1048, 1083 (C.D. Cal. 2012), aff’d in part, 642 F. App’x 680 (9th Cir.

2016); see also FTC v. Johnson, 96 F. Supp. 3d 1110, 1142 (D. Nev. 2015) (rejecting “fair notice” argument as to

Section 5(a) liability and noting there is “extensive case law and guidance on what constitutes a deceptive act or

practice under Section 5(a)”).

of products and services, 28 including new and evolving technologies. 29 And we have taken aim at

companies that failed to provide adequate disclosures, 30 as well as governmental and private

standard-setting entities that sought to restrict truthful advertising. 31

27F

28F

29F

30F

IV.

Deceiving Consumers as to the Objectives of an AI Systems Is a Section 5 Violation

AI products and services are not exempt from Section 5’s long-established, and generally

applicable reach. The Commission has already brought a number of recent enforcement actions to

combat deceptive claims in connection with AI-powered scams, as well as material

misrepresentations about the performance, efficacy, and characteristics of AI products. 32 For

instance, the Commission recently invoked well-established Section 5 principles to address an AI

company’s misleading claims about the ability of its conversational AI tool to replace human

customer service representatives. 33 Although the Commission is careful to avoid unduly burdening

innovation in the AI industry, 34 the Commission will continue to protect consumers and the market

31F

32F

33F

See, e.g., FTC v. Fleetcor Techs., Inc., 620 F. Supp. 3d 1268, 1289–1313 (N.D. Ga. 2022) (fuel cards and

discounts), aff’d sub nom FTC v. Corpay, Inc., 164 F.4th 807 (11th Cir. 2026); FTC v. D-Link Sys., Inc., No. 3:17CV-00039-JD, 2017 WL 4150873, at *2 (N.D. Cal. Sept. 19, 2017) (data security and protections for IoT devices);

Pom Wonderful, 777 F.3d at 490–500 (dietary supplements); Com. Planet, Inc., 878 F. Supp. 2d at 1083 (online

webpages and negative option plans).

See, e.g., Stipulated Order, FTC v. Amazon.com, Inc., No. 2:23-CV-00932-JHC (W.D. Wash. Sept. 25, 2025), Dkt.

No. 535 (enrolling consumers into online shopping memberships without consumers’ express informed consent);

Compl., United States v. Facebook, No. 19-cv-2184 (D.D.C. July 24, 2019), Dkt. No. 1 (misrepresenting that users

would have to “turn[] on” facial-recognition technology); Compl., FTC v. New Consumer Sols., LLC, No. 1:15-cv01614 (N.D. Ill. Feb. 23, 2015), Dkt. No. 1 (misrepresenting ability to detect melanoma by analyzing pictures of

consumers’ skin); Compl., In re Sears Holdings Mgmt. Corp., FTC Docket No. C-4264 (Sept. 9, 2009) (tracking

consumers’ “online browsing”).

See, e.g., FTC v. LendingClub Corp., No. 18-CV-02454-JSC, 2020 WL 2838827, at *6 (N.D. Cal. June 1, 2020)

(inadequate disclosures of loan origination fees); see also FTC v. Cyberspace.Com LLC, 453 F.3d 1196, 1200 (9th

Cir. 2006) (“A solicitation may be likely to mislead by virtue of the net impression it creates even though the

solicitation also contains truthful disclosures.”).

See, e.g., In re Am. Med. Ass’n, 94 F.T.C. 701, 235 (1979) (restricting physician advertising).

See, e.g., Consent Order, In re Workado, LLC, No. C-4822 (F.T.C. Aug. 21, 2025) (deceptive claims regarding

accuracy or efficacy of its AI-content detection products),

https://www.ftc.gov/system/files/ftc_gov/pdf/ContentatScaleAI-DecisionandOrder.pdf; Stipulated Order, FTC v.

TheFBAMachine Inc., No. 24-cv-6635-JXN-LDW (D.N.J. July 31, 2025), Dkt. No. 162 (falsely guaranteeing

consumers could make money operating online storefronts using AI-powered software); Consent Order, In re

DoNotPay, Inc., FTC Docket No. C-4812 (Jan. 14, 2025) (deceptive claims about the ability of its AI chatbot to

replace the services of a human lawyer),

https://www.ftc.gov/system/files/ftc_gov/pdf/2323042_donotpay_decision_and_order_0.pdf; Consent Order, In re

Intellivision Techs. Corp., No. C-4809 (F.T.C. Jan. 8, 2025) (deceptive claims regarding accuracy or efficacy of its

AI-powered facial recognition software),

https://www.ftc.gov/system/files/ftc_gov/pdf/2323023c4809intellivisionfinalorder.pdf; Stipulated Order, FTC v.

Evolv Techs. Holdings, Inc., No. 1:24-cv-12940-PGL (D. Mass. Nov. 26, 2024), Dkt. No. 2 (false claims about the

efficacy of its AI-powered security screening system for detecting weapons).

See Compl., FTC v. Air Ai Techs., Inc., No. 2:25-cv-03068-SMB (D. Ariz. Aug. 25, 2025), Dkt. No. 1 (deceptive

claims regarding efficacy and profitability related to the company’s “conversational AI” product).

In fact, the Commission has already taken steps to address these unnecessary burdens, for example, by setting

aside the previous Commission’s order against a generative AI company. See Order Reopening and Setting Aside

Order at 5, In re Rytr LLC, FTC Docket C-4806 (Dec. 22, 2025), https://www.ftc.gov/system/files/ftc_gov/pdf/Rytr28

from deceptive or unfair business practices, including those relying on AI systems or technologies,

by using well-established Section 5 principles.

In marketing their products as problem-solving tools, AI companies have represented

explicitly and implicitly that their AI systems aim to produce the best output possible given

technological and resource constraints. 35 AI companies aggressively market themselves as

building products and services that distill all of human knowledge to solve problems, large and

small, in furtherance of users’ given objectives. 36 This is because companies know that their

success depends on their products’ utility, and an AI system’s utility is a function of its ability to

accomplish the objectives that users ask it to accomplish. 37

34F

35F

36F

Consumers share this understanding of the purpose of AI systems and have learned to trust

these systems across a broad swath of applications, from education to health, finance, relationships

Order.pdf (“Where actors use AI to violate the law or deceive consumers about the capabilities of their generative

AI, they should be held accountable, as the FTC has done and will continue to do. But that is not the case here.

Rushing in to impose aggressive law enforcement unsupported by facts or law is improper and is not in the public

interest.”).

See, e.g., OpenAI, Why Language Models Hallucinate (Sept. 5, 2025) (“At OpenAI, we’re working hard to make

AI systems more useful and reliable. Even as language models become more capable, one challenge remains

stubbornly hard to fully solve: hallucinations.”), https://openai.com/index/why-language-models-hallucinate/;

Anthropic, Claude Product Overview, https://claude.com/product/overview (last visited May 22, 2026) (“Meet your

thinking partner … Tackle any big, bold, bewildering challenge with Claude … The AI for problem solvers …

Tackle your toughest work … Like an expert in your pocket … There’s never been a worse time to be a problem, or

a better time to be a problem solver … What problem are you up against?”); Anthropic, How People Ask Claude for

Personal Guidance, https://www.anthropic.com/research/claude-personal-guidance (Apr. 30, 2026) (“Helpfulness is

one of Claude’s most important traits. Speaking with Claude should be akin to a conversation with a brilliant friend,

one who will speak frankly to a person about their situation, providing information grounded in evidence.”); X.ai,

Chat With Grok, https://x.ai/grok (last accessed May 20, 2026) (“Grok is your truth-seeking AI companion for

unfiltered answers with advanced capabilities in reasoning, coding, and visual processing.”).

See, e.g., Anthropic, Claude, https://claude.com (last visited May 22, 2026) (“What is Claude and how does it

work? Claude is an artificial intelligence, trained by Anthropic using Constitutional AI to be safe, accurate, and

secure — the trusted assistant for you to do your best work. … What should I use Claude for? If you can dream it,

Claude can help you do it. Claude can process large amounts of information, brainstorm ideas, generate text and

code, help you understand subjects, coach you through difficult situations, simplify your busywork so you can focus

on what matters most, and so much more.”); OpenAI, ChatGPT Capabilities Overview

https://help.openai.com/en/articles/9260256-chatgpt-capabilities-overview (last visited Mar. 17, 2026) (touting

ChatGPT’s ability to solve problems); OpenAI, Extracting Insights with ChatGPT Data Analysis,

https://help.openai.com/en/articles/9213685-extracting-insights-with-chatgpt-data-analysis (last visited Mar. 17,

2026) (claiming that ChatGPT can identify missing or outlier data); OpenAI, ChatGPT for Healthcare,

https://help.openai.com/en/articles/20001046-chatgpt-for-healthcare (last visited Mar. 17, 2026) (“ChatGPT for

Healthcare can pull from millions of peer-reviewed studies, clinical guidelines, and public health sources” to “[g]et

clinical answers with citations” that “perform[] better than human baselines across every role measured”).

Similarly, an ability to make accurate predictions, correctly solve problems, or make good decisions hinges on AI

systems operating without undisclosed ideological biases that contradict explicit and implicit claims of accuracy and

neutrality. Though the exact line of what constitutes bias may be difficult to draw, forcing an AI system to prioritize

a singular ideological objective over all else is on the wrong side. See Concurring Statement of Comm’r Andrew N.

Ferguson, In re IntelliVision Techs. Corp., Matter No. 2323023, at 1 (F.T.C. Dec. 3, 2024),

https://www.ftc.gov/system/files/ftc_gov/pdf/intellivision-ferguson-concurrence.pdf.

or a myriad of other uses. 38 In fact, according to one major AI developer, consumers accept AI

system outputs without conducting any further fact checking over 90% of the time, even though

these systems purport to only strive for, rather than guarantee, complete accuracy. 39 In other words,

consumers ask AI systems for help answering deeply personal and important questions, and they

overwhelmingly trust that the system is designed to provide an answer tailored specifically to their

objectives. 40 And because of both the companies’ marketing and the inherent value proposition of

an AI system, this widely held consumer expectation is eminently reasonable.

37F

38F

39F

Of course, AI systems are likely to simultaneously pursue multiple objectives, some

explicit and some implicit. This will often be consistent with consumers’ reasonable expectations.

Users of an AI chatbot might, for example, reasonably expect the system to balance succinctness,

clarity, relevance, accuracy, and other objectives in its attempt to produce the best output. And

while truth and accuracy are in many cases implicit objectives in requests to AI systems, a user

could request output that is intentionally inaccurate or that deprioritizes accuracy in favor of some

other objective like entertainment.

These representations and baseline consumer expectations notwithstanding, AI companies

might steer their systems’ outputs toward objectives other than those that consumers ask for or

expect. A company could be tempted, for example, to abuse consumer trust by training a model

surreptitiously to produce ideologically motivated distortions in a response to a factual question,

such as to correct what the developer believes are “historical injustices” in the facts. 41 Or a

company could be pressured by a state law to alter its technology’s outputs. The original version

of Colorado’s Artificial Intelligence Act, 42 for example, imposed a broad duty on AI companies to

avoid output that might lead to disparate impacts in various contexts, including when a customer’s

foreseeable use of that output could itself create a disparate impact. And the revised version of that

law explicitly provides that AI companies can be held liable for discriminatory outcomes caused

40F

41F

Paulo Vargas, Your Teen Is Probably Using AI for Homework, Digital Trends (Mar. 2, 2026) (“More than half of

U.S. teens ages 13 to 17 say they have turned to chatbots . . . for school tasks.”),

https://www.digitaltrends.com/computing/your-teen-is-probably-using-ai-for-homework/; Walt Williams, Survey:

Consumers Increasingly Turn to AI for Financial Advice, ABA Banking Journal (Sept. 2, 2025) (“A majority of

consumers are turning to artificial intelligence for financial advice . . . .”),

https://bankingjournal.aba.com/2025/09/survey-consumers-increasingly-turn-to-ai-for-financial-advice/; Theo

Burman, Americans Are Using AI To Diagnose Their Health Issues, Newsweek (July 20, 2025) (“Both clinicians

and patients are using artificial intelligence more and more to help diagnose illness and injuries . . . .”),

https://www.newsweek.com/ai-healthcare-diagnosis-chatgpt-doctor-2100091; How People Ask Claude for Personal

Guidance, Anthropic (Apr. 30, 2026) (finding that 12% of all conversations with the Claude program concerned

relationship navigation, 6% concerned personal development, and 4% concerned spirituality),

https://www.anthropic.com/research/claude-personal-guidance.

Ted Ladd, Anthropic: 91% of Users Do Not Fact-Check AI. Let’s Fix That, FORBES (Mar. 5, 2026),

https://www.forbes.com/sites/tedladd/2026/03/05/anthropic-91-of-users-do-not-fact-check-ai-lets-fix-that/.

Reasonable consumers would not expect, of course, that an AI system would output content that is clearly and

unequivocally not protected by the First Amendment, such as child pornography. And nothing in this statement shall

be construed to prohibit companies from imposing limits on model use to prevent cybersecurity attacks.

See, e.g., Leo Briceno, Anthropic’s Moral Compass Architect Suggested AI Overcorrection Could Address

Historical Injustices, Fox News (Apr. 22, 2026), https://www.foxnews.com/politics/anthropics-moral-compassarchitect-suggested-ai-overcorrection-could-address-historical-injustices.

Colo. Sen. Bill 24-205 (enacted May 17, 2024), repealed and reenacted, as amended, by Colo. Sen. Bill 26-189

(enacted May 14, 2026).

by their customers’ use of their products. 43 It is predictable that an AI company might suppress

accuracy and interpose other objectives, such as so-called “equity,” to avoid liability under this

law, but fail to disclose these ulterior objectives in order to hide the loss of accuracy they

necessitate. 44 AI companies may also be subject to public pressure to surreptitiously avoid

politically inflammatory outputs, as well as the temptation to clandestinely modify their AI systems

to further their own or their employees’ political agendas.

42F

43F

In light of the above, the 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 45 consumers in violation of Section 5 of the FTC Act. 46

AI companies have made both explicit and implicit representations that their systems aim to

produce outputs that achieve users’ objectives as faithfully and accurately as they can. 47 Those

representations, absent adequate disclaimers or qualifications, are material as consumers rely on

them and are likely to consider, in their decision as to which AI system to use or pay for, whether

a particular system is designed to produce output accomplishing their objectives or to intentionally

produce a worse output in service of another objective. Consumers may be induced to pay for a

service that does not behave as advertised. Consumers may also be deceived into relying on a

technology that, by design, produces worse outputs and may recommend suboptimal courses of

action, not because of any technological or resource limitations, but because the AI developer’s

hidden agenda subverted consumers’ objectives.

44F

45F

46F

As is always the case, a company’s motives for deceiving consumers are irrelevant to the

application of Section 5. 48 Whether motivated by profit, shaping public opinion, or anything else,

Section 5 prohibits deceiving consumers. These prohibitions apply even when a company engages

in a deceptive act or practice in order to comply with a state law. Although the FTC Act does not

expressly preempt state law, state law is impliedly preempted to the extent it conflicts with a federal

47F

Colo. Sen. Bill 26-189, 6-1-1707 (enacted May 14, 2026).

See Exec. Order No. 14281, Restoring Equality of Opportunity and Meritocracy, at § 1, 90 Fed. Reg. 17537 (Apr.

23, 2025) (“Disparate-impact liability all but requires individuals and businesses to consider race and engage in

racial balancing to avoid potentially crippling legal liability. . . . On a practical level, disparate-impact liability has

hindered businesses from making hiring and other employment decisions based on merit and skill, their needs, or the

needs of their customers because of the specter that such a process might lead to disparate outcomes, and thus

disparate-impact lawsuits. . . . Disparate-impact liability imperils the effectiveness of civil rights laws by mandating,

rather than proscribing, discrimination.”). Indeed, it seems likely that a law restricting truthful speech because it

might lead another person to commit disparate impact discrimination would not survive First Amendment scrutiny,

but that is orthogonal to the FTC Act issues we address here.

The Commission at this time takes no position on whether the practices discussed in this statement may also be

unfair under the FTC Act.

This conduct is distinct from the problem of incorrect output from AI systems, frequently called “hallucinations,”

that stem not from a design decision to prioritize objectives contrary to users’ reasonable expectations, but from the

technological and resource limitations that AI systems necessarily reflect. While a company may, for example,

unlawfully deceive consumers if it misrepresents the likelihood of such hallucinations, the Commission does not

believe that such hallucinations in and of themselves raise issues under Section 5 or any other law that the

Commission enforces.

See supra notes 35 & 36.

FTC v. Bay Area Bus. Council, 423 F.3d 627, 635 (7th Cir. 2005) (deception); In re Kraft, Inc., 114 F.T.C. 40, 122

(1991) (deception).

regulatory scheme. 49 A state law that requires an AI firm to deceive its consumers obviously

conflicts with Section 5’s express purpose of protecting consumers from such conduct.

48F

An AI company can, of course, take actions that shape consumer expectations in ways that

make it unreasonable for consumers to believe the AI system is designed to achieve the objectives

that consumers would otherwise expect. A company can, for example, clearly and conspicuously

disclose that its systems are designed to produce outputs that prioritize certain objectives over what

users request and otherwise expect. But such a disclaimer would have to be adequate to shift

consumer expectations that would be based otherwise both on companies’ explicit representations

and on the inherent value proposition of AI systems as tools to solve human problems. An adequate

disclaimer could not be buried in terms of service, for instance. It would have to clearly and

conspicuously dispel the notion that the system is designed to give the best answer possible. Such

a disclaimer would need to be prominent, and it is doubtful that a one-time disclosure subsequently

hidden away in fine print would suffice. The more that the disclosure cuts against the reasonable

expectations that users would take away from other contexts, the more persistent and prominent it

would need to be. 50 A prominent misrepresentation is unlikely to be remedied by a less prominent,

subsequent disclosure. 51 Ultimately, it is the company’s responsibility to ensure compliance with

Section 5.

49F

50F

Schneidewind v. ANR Pipeline Co., 485 U.S. 293, 300 (1988) (“[S]tate law is pre-empted when it actually

conflicts with federal law. Such conflict will be found when it is impossible to comply with both state and federal

law, or where the state law stands as an obstacle of the full purposes and objectives of Congress.” (internal

quotations omitted)).

FTC v. Direct Mktg. Concepts, Inc., 624 F.3d 1, 12 (1st Cir. 2010) (explaining that disclaimers “are not adequate

to avoid liability unless they are sufficiently prominent and unambiguous to change the apparent meaning of the

claims and to leave an accurate impression”)); Removatron Int'l Corp. v. FTC, 884 F.2d 1489, 1497 (1st Cir.1989)

(“[d]isclaimers or qualifications in any particular ad are not adequate to avoid liability unless they are sufficiently

prominent and unambiguous to change the apparent meaning of the claims and to leave an accurate impression.”).

See, e.g., Deception Policy Statement at *48, supra note 18 (“Depending on the circumstances, accurate

information in the text may not remedy a false headline because reasonable consumer may glance only at the

headline.”); see also FTC v. Corpay, Inc., 164 F.4th 807, 836 (11th Cir. Jan. 6, 2026) (disclaimers or qualifying

language may not correct a misleading impression when the disclaimer “is small, ambiguous, or contradicted by the

body of the ad”) (citing FTC v. Cyberspace.com LLC, 453 F.3d 1198, 1200-01 (9th Cir. 2006); FTC v. Direct Mktg.

Concepts, Inc., 624 F.3d 1, 12 (1st Cir. 2010) (explaining that disclaimers “are not adequate to avoid liability unless

they are sufficiently prominent and unambiguous to change the apparent meaning of the claims and to leave an

accurate impression”); Removatron Int'l Corp. v. FTC, 884 F.2d 1489, 1497 (1st Cir.1989) (“[d]isclaimers or

qualifications in any particular ad are not adequate to avoid liability unless they are sufficiently prominent and

unambiguous to change the apparent meaning of the claims and to leave an accurate impression.”).

This is a copy of a public record, reproduced as it was published. It is not legal advice, and it may not be the version a court would rely on. Check the official source before you cite it.