Nice on paper: a regulator finally grabs hold of the question of whether 'tuning' a model's output is deceptive. For now it's only a proposal - but the principle is clean.
Fact
The facts: on July 1, 2026 the FTC proposed a policy statement, 'Suppression of Accuracy in Artificial Intelligence Systems' (published in the Federal Register on July 7; docket FTC-2026-0859). The thesis: if a company presents its product as giving the most accurate answer, but secretly steers the output toward another goal (say, ideological) without clear and visible disclosure, that can be fraud under Section 5 of the FTC Act (the part of the law that bans deceptive practices). The statement draws a line between deliberate steering and 'hallucinations', which come from technical limitations and are not by themselves a violation. Vote 2-0; public comment period through July 31, 2026. Primary source: FTC (ftc.gov), Federal Register.
Let's not blow this out of proportion - it's the US, and for now it's only a proposal, not a law. But the principle lines up with our own: honesty about what your system actually does. Disclosure, not hidden steering.
If you build on AI, say plainly what it's optimizing for. Hidden steering hurts twice: it kills the user's trust and opens a legal door you don't want open. Disclose it up front and you stay clean; hide it and you're betting your reputation - and leaving your back exposed - on nobody ever pulling the curtain.
Sources
FTC - Seeks Public Comment on AI Accuracy Policy StatementFederal Register - Suppression of Accuracy in AI SystemsThe visual is generated code-art, no third-party images.