top of page
Search

The FTC's AI "Accuracy" Gambit — and the 109 State Laws It's Trying to Outrun


On July 1, the Federal Trade Commission published a proposed policy statement with a dry title and a sharp legal edge: "Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems." The Commission voted 2-0 to put it out for public comment, and the core argument is straightforward. When an AI company markets its system as accurate, objective, or unbiased, that claim creates an expectation. If the company then quietly tunes the model's outputs toward some other, undisclosed goal — an ideological preference, a business interest, or, notably, compliance with a state law — the mismatch between the promise and the product can itself be a deceptive act under Section 5 of the FTC Act.

The statement is careful to draw a line: it explicitly excludes ordinary hallucinations, the wrong answers that come from a model's technical limits, since those aren't a deliberate design choice. What it targets is intentional steering. And it gives companies an out — they can prioritize something other than raw accuracy as long as they disclose that trade-off clearly, prominently, and persistently, not buried in a terms-of-service paragraph nobody reads.

The part that turns this from a compliance memo into a genuine legal fight is who else the FTC named. The statement calls out Colorado's Artificial Intelligence Act by name, arguing that a state law requiring companies to adjust outputs to avoid "algorithmic discrimination" could itself force the kind of undisclosed manipulation the FTC considers deceptive — meaning a company complying with Colorado's law in good faith could still be violating federal law by not disclosing that it did so. The FTC's conclusion: state requirements that conflict with this reading of Section 5 are "impliedly preempted." This wasn't a random legal theory. The statement was issued under a direct instruction in Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," which President Trump signed on December 11, 2025 — an order that specifically told the FTC to address how state laws requiring altered AI outputs conflict with federal law.

Here's the part that makes this a story about more than one agency's legal theory: the states didn't wait to find out how it plays out. According to a count published by NYU's Center on Technology Policy in Tech Policy Press, U.S. states enacted 109 AI laws and 28 data-center statutes across 29 states in just the first half of 2026 — this despite a December 2025 executive order that created a federal task force specifically to challenge state AI laws in court. Coverage of the FTC statement's rollout noted that California and New Jersey passed new AI laws in the days immediately around it, seemingly unbothered by Washington's preemption claim. Congress, meanwhile, is trying a third approach entirely: a bipartisan House draft nicknamed the "Great American AI Act" would trade a three-year freeze on new state AI-development laws for a federal disclosure framework administered by the Commerce Department, and Senator Ed Markey unveiled a competing package of roughly a dozen bills on July 10 covering data-center certification, workplace automation, child safety, healthcare AI, and algorithmic bias.

So, at the same moment, one federal agency is asserting it can override state AI law through a legal theory that hasn't been tested in court, a separate federal task force exists to sue states directly, Congress has at least two competing bills proposing to settle the question legislatively, and states are still passing new laws as if none of that is happening. None of it is final. The FTC's statement is open for public comment through July 31, 2026, and isn't a rule yet. But for anyone deploying AI in a real organization, "wait for it to resolve" is not a strategy — because there's no clear timeline for when, or whether, it resolves at all.

For organizations trying to use AI responsibly right now, this raises a few concrete questions:

  • If you operate across multiple states, whose AI rules actually govern your deployment today — and does that answer change if the FTC's preemption theory is later upheld, narrowed, or struck down in court?

  • If a vendor tunes a model's outputs to comply with a state law like Colorado's, are you now expected to independently verify that they disclosed that adjustment clearly enough to satisfy the FTC's reading of Section 5?

  • Should your organization build compliance processes to the strictest applicable state standard as insurance, even while a federal agency is actively arguing that standard may not survive?

  • How do you document your own model's known behavior and disclosures now, before regulatory clarity arrives, so you're not caught flat-footed by whichever rule wins?

At OFER AI, this is exactly the kind of story our forums exist to unpack — not to predict how the legal fight ends, but to help people build practices that hold up regardless of the outcome. A regulatory environment this unsettled is precisely why we've been developing a model-vetting rubric that treats governance, security, and cost controls as living categories to revisit, not one-time checkboxes — because "we followed the rules" is a moving target when the rules themselves are in open dispute between the executive branch, Congress, and 29 state legislatures.

In upcoming deep dives, we'll be exploring:

  • How to build an internal disclosure and documentation practice that would satisfy either a strict-state-law standard or the FTC's "clear, conspicuous, persistent" disclosure test, so you're covered under both.

  • What "reasonable diligence" looks like when evaluating a vendor's compliance claims in a regulatory environment where federal and state authority are actively contesting each other.

  • How multi-state organizations should be tracking this patchwork on an ongoing basis, rather than waiting for a single definitive ruling that may not arrive this year.

If your organization has already had to make a call on this — picking a compliance posture before the dust settles or fielding a client question about which state's AI rules actually apply — that's exactly the kind of real-world decision we want in our Notes from the Field segment. These lived examples are what turn a legal news story into something the rest of the community can actually use.


Sources

Postscript — related podcasts

  • The Framework Podcast, episode of July 7, 2026 — covers the FTC statement's timing relative to the frontier-AI executive order deadline and California/New Jersey's new AI laws in the same news cycle. Listen here.

  • Yesterday in AI, episode "The Email That Blew Up Anthropic's Pentagon Deal and the Billion-Dollar Pivot on AI Layoffs" (July 7, 2026) — includes a segment breaking down the FTC's Section 5 theory and what it means for companies that tune live models. Listen here.

Comments

Rated 0 out of 5 stars.
No ratings yet
Add a rating

Join us on mobile!

Download the “” app to easily stay updated on the go.

Scan QR code to join the app

SUBSCRIBE & JOIN

Sign up to receive Open Forum news and updates.

Subscribing to our newsletter is free of charge and notifies you of new blog posts, upcoming events and new online programs.  Becoming a member provides you with other benefits.

SCROLL

Becoming a member is free of charge and gives you access to additional content, the ability to register for in-person and online events as well as online programs.  Members can participate in roundtable discussions, deep dives and be heard. Tiered plans are only available to site members. 

​

Become part of the AI Solution.  Join Now.

bottom of page