The Federal Trade Commission wants to use existing consumer protection law against artificial intelligence companies that present their systems as accurate but deliberately steer answers toward another goal without clearly telling users.
The proposal, published July 1 under Section 5 of the FTC Act, calls this “suppression of accuracy” and treats it as a form of consumer deception, according to the FTC’s policy statement. PYMNTS reported on the FTC’s call for comments when it opened in July. Companies can avoid a deception claim by clearly disclosing when their AI prioritizes goals other than what users request or reasonably expect.
The FTC draws a specific line between this kind of deliberate steering and a hallucination, when AI gives a wrong answer because of its own technical limitations. Hallucinations and other mistakes caused by the technology itself don’t count as suppression of accuracy on their own, the FTC said in its proposal. The FTC’s target is narrower: a company that configures its AI to favor a different goal over accuracy on purpose, without disclosing that fact to users.
That distinction could be difficult to enforce, according to a legal analysis by Covington & Burling’s Inside Privacy blog. Large language models can give different answers to the same prompt, and there isn’t always a single correct output to compare against. The FTC’s nine-page proposal offers no technical test for determining whether an inaccurate answer was a hallucination or was deliberately changed. That could put internal model tests, system instructions and records showing how developers changed model behavior at the center of future cases.
A Different Approach Than Other AI Rules
The FTC’s approach also differs from how other jurisdictions have handled AI accuracy. The European Union’s AI Act requires certain high-risk systems to meet an accuracy standard and disclose those limitations directly in the system’s instructions for use, according to Article 15 of the law. U.S. state laws have mostly focused elsewhere, targeting discrimination and high-risk decisions like hiring or lending rather than accuracy itself.
The FTC is asking a narrower question than either approach: whether deliberately steering an AI model away from accuracy can itself constitute consumer deception. The proposal also says complying with a state artificial intelligence rule would not necessarily shield a company from federal enforcement. That means a company that adjusts an otherwise accurate output to satisfy a state requirement could still face an FTC deception claim if that change wasn’t clearly disclosed to users.
The Policy Has Not Been Adopted
Public comments on the proposal closed July 31, and the FTC has not said whether or when it plans to finalize the policy. That leaves companies in an uncertain position: the FTC’s own framework already describes what it expects, disclosing clearly whenever an AI system prioritizes something other than pure accuracy. But nothing in the current proposal carries the force of law until the agency takes a further step.
Whether the FTC moves forward with the policy largely as written, changes it based on the comments it received, or lets it remain an unadopted statement of intent is still an open question. So is whether “suppression of accuracy” becomes an actual enforcement standard, or just a record of how the current FTC thinks about AI accuracy.
The post The FTC Is Coming for AI’s Honesty Problem appeared first on PYMNTS.com.