AI Regulation Tracker / Regulation proposed
FTC Proposes Treating Suppressed AI Accuracy as a Deceptive Act Under Section 5
The FTC has published a proposed policy statement warning that companies that degrade or suppress an AI system's accuracy while marketing it as accurate or objective may be committing a deceptive act. Comments are due July 31, 2026.
The Federal Trade Commission on July 7, 2026 published a proposed policy statement that puts a specific new label on an old idea: telling customers your AI is accurate, then quietly making it less so, can be a deceptive act. The document, "Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems," appears in the Federal Register at 91 FR 41638 (document 2026-13628) as a request for comments. It is not a final rule. Comments are due on or before July 31, 2026, through docket FTC-2026-0859 on regulations.gov.
What the FTC is proposing
The Commission proposes to apply the prohibition on deceptive acts or practices in Section 5 of the FTC Act to companies that market AI systems. The concern it names is narrow and concrete: businesses that degrade or suppress the accuracy of AI outputs, including to serve undisclosed objectives, while representing those systems as accurate, objective, or reliable. In the FTC's framing, the mismatch between the representation and the product is where the deception lives. If a company tells users its system is objective and unbiased, and then knowingly manipulates the output in a way that contradicts that promise, the Commission's position is that Section 5 can reach the conduct.
Who carries the duty
The duty attaches to the party making the claim. Any business that markets or represents an AI system's outputs as accurate, objective, or unbiased is on the hook for that representation. That reaches beyond model developers. A professional who licenses a third-party AI tool and then describes it to clients as reliable has made a representation of their own. The practical standard the proposal implies is substantiation: if you say the output is accurate, you should be able to support that claim, and you should not overstate what the system can do.
The phrase the Commission keeps returning to is undisclosed objectives. The worry is not that AI systems make mistakes, since every system has an error rate. The worry is a company that markets a system as neutral while tuning it toward a hidden commercial or editorial goal, then leaves the buyer believing the output is objective. In that scenario the accuracy claim and the private steering point in opposite directions, and the customer never learns of the gap. For professionals who pass AI output to clients, the lesson is that silence about how a tool is configured can be as much a problem as an affirmative false statement.
What it does not do
The proposed statement does not create a new statute, and it does not, by itself, ban any product or feature. As a proposed policy statement open for public comment, it is not yet operative and its final language could change. It does not set numeric accuracy thresholds or certification requirements. What it does is signal how the Commission intends to read existing law. That distinction matters, because Section 5 deception liability already exists independently of this statement. A firm that misrepresents the reliability of an AI system can face that theory today, whether or not this particular policy statement is finalized.
Why professionals should read it now
For licensed practitioners, the useful move is defensive documentation. If your client-facing materials call an AI tool accurate, objective, or unbiased, treat that as a claim you may have to defend. Keep a written record of the system's known limitations, avoid marketing language that promises more than the tool delivers, and be careful about configuring or steering outputs in ways that would contradict what you have told users. None of this requires waiting for the comment period to close. The reasonable-substantiation instinct behind the proposal is the same one that already governs any factual claim a business makes about its products.
There is also a plain drafting exercise worth doing now. Review the words your firm actually uses about the AI it relies on. Terms like accurate, objective, unbiased, and reliable carry weight under this proposal, while calibrated language, such as a description of what the tool is designed to do and where a human still reviews the result, carries far less risk. The goal is not to hide the technology but to describe it truthfully, so the representation and the product match. A short internal review of contracts, websites, and sales scripts can surface the overclaims before a regulator or a client does.
The comment window is short. Firms that deploy or resell AI output, and trade groups that represent them, have until July 31, 2026 to weigh in through the FTC's docket.
Frequently Asked Questions
What exactly did the FTC change on July 7, 2026?
It published a proposed policy statement (91 FR 41638) saying it will treat the suppression or degradation of an AI system's accuracy as a possible deceptive act under Section 5 when the company markets that system as accurate, objective, or reliable. It is a request for comments, not a final rule.
Who does this affect?
Any business that markets or represents an AI system's outputs as accurate, objective, or unbiased. That includes AI developers and also professionals or firms that deploy third-party AI and describe it to clients as reliable.
Is this in force now?
No. It is a proposed policy statement open for public comment through July 31, 2026, so it is not yet operative. But Section 5 deception liability already exists on its own, so misrepresenting an AI system's reliability can be actionable regardless.
What should a professional who uses AI tools do about it?
Be able to substantiate any accuracy or objectivity claim you make about the tool, document its known limitations, avoid overstating reliability, and do not manipulate outputs in ways that contradict what you have told clients. If it affects your practice, file a comment before July 31, 2026.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.