Compliance Looks Like Fraud to the FTC

The Federal Trade Commission just proposed that steering an AI system's outputs to satisfy state law — without telling users — is likely federal fraud. That turns every vendor selling into hiring, lending, or housing into a steer-or-disclose trap: comply with Colorado-style rules and risk Section 5, or disclose the bias filter and watch enterprise buyers walk.

What's new: On July 7, the FTC published a proposed policy statement in the Federal Register (comments due July 31) applying Section 5 of the FTC Act to AI vendors that alter model outputs away from what users reasonably expect. Chairman Andrew N. Ferguson said the FTC wants to hear from businesses and consumers "about their experiences and concerns regarding the subversion of AI systems for ideological ends." Steering done to comply with state law does not get a safe harbor.

Why now:

• Executive Order 14365 (December 11, 2025) ordered the FTC to explain how state laws requiring output changes conflict with federal consumer protection.
• On June 29, the Supreme Court held in Trump v. Slaughter that the president may remove FTC commissioners at will — ending 90 years of Humphrey's Executor independence.
• Two weeks later, the FTC filed the steer-or-disclose theory anyway.

The Colorado irony:

• Colorado's original S.B. 24-205 imposed a duty to prevent "algorithmic discrimination" in consequential decisions — employment, housing, lending, health care.
• xAI sued April 9; the Justice Department intervened April 24; enforcement was suspended April 27.
• Governor Jared Polis signed S.B. 26-189 on May 14 — repealing the algorithmic-discrimination duty and replacing it with notice, documentation, and fault allocation under existing anti-discrimination law. Effective date: January 1, 2027.
• The FTC proposal still cites S.B. 26-189 § 6-1-1707 — arguing vendors can be held liable for discriminatory outcomes caused by customers' use, and may secretly steer outputs toward "equity" to avoid that exposure.

What the FTC says is deception:

• Training or tuning models to advance undisclosed ideological goals.
• Altering outputs under political or public pressure without disclosure.
• Steering to avoid state-law liability when users were sold "truthful and accurate" problem-solving.
• Burying the real objective in terms-of-service fine print.

The preemption play: The FTC Act has no express state-law safe harbor. The proposal argues state laws that force vendors to steer outputs away from user expectations are impliedly preempted when they conflict with Section 5's consumer-protection purpose. Enforcement would target AI companies — not state regulators.

What practitioners said:

• Tyler Thompson of Reed Smith told Stateline (July 10): "Just the fact that companies could be tweaking their models and that could lead to a deceptive trade practice, I think is huge news."
• Noah Kenney of Digital 520 told Stateline the statement is "signaling and pressure, not legal preemption" — especially paired with the December executive order's DOJ AI litigation task force.
• Kenney also flagged the mirror risk: "A federal effort to dictate what counts as a 'neutral' or 'accurate' output raises its own First Amendment concerns about compelled speech."

The numbers:

• 1,000+ AI bills introduced in state legislatures, per the White House fact sheet cited in the Federal Register.
• 91% of Claude users accept outputs without fact-checking, per a Forbes piece the FTC cited — making undisclosed steering material to consumer reliance.

Who's squeezed: Developers and deployers tuning models for fair-lending, hiring-screening, or benefits decisions in Colorado, California, or the dozen states drafting similar rules. Labs marketing "truth-seeking" chatbots while running undisclosed safety or compliance layers face the sharpest exposure.

Convina's view: Washington is fighting the last war. Denver already dropped the algorithmic-discrimination duty that triggered xAI's suit — yet the FTC is still writing enforcement theory against steering for equity. That is not neutral deregulation; it is a national steer-or-lie standard handed to a commission the president can now fire at will. Vendors will not stop tuning models — they will stop admitting it. The compliance stack just became a disclosure game, and the only winners are lawyers billing both sides of the preemption fight.