The Order Meant to Sideline State AI Laws Came Back as a Disclosure Form

On December 11, 2025, President Trump signed Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," directing the Federal Trade Commission to use Section 5 of the FTC Act - the general federal ban on unfair or deceptive practices - against AI companies whose systems produce outputs shaped to satisfy state law rather than user expectations. The order gave the Commission 90 days to say how. The FTC's answer arrived on July 7, 2026, roughly four months late, and it does not do what the order was written to do. It does not preempt the state laws Executive Order 14365 singles out. It proposes a new federal condition layered on top of them: comply with a state law's output-alteration requirement quietly, the FTC's draft says, and the silence - not the compliance itself - is what could now violate federal law.

What the order actually asked for

Executive Order 14365's Section 1 purpose statement names its target directly: state laws like Colorado's ban on "algorithmic discrimination," which the order says "may even force AI mo[dels] to produce false results." Section 7 translates that objection into an instruction: within 90 days, the FTC Chairman was to issue a policy statement on how Section 5's deceptive-practices prohibition applies to AI models, and specifically to address when state laws requiring "alterations to the truthful outputs of AI models" conflict with that federal prohibition.

The order was not a blanket assault on state AI regulation. Section 8, which separately directs preparation of legislation for a uniform federal framework, carves out state laws on child safety protections, AI compute and data-center infrastructure (aside from generic permitting rules), and state government procurement and use of AI. Whatever Congress is eventually asked to preempt, those three categories were excluded from the start. The FTC's assignment sat inside a narrower target: private-sector AI systems and the outputs they show ordinary users.

What the FTC delivered - four months late

The Commission's proposed policy statement, published in the Federal Register as "Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems," states plainly that it exists because "the Executive Order directs the Commission to issue this enforcement policy statement clarifying the application of section 5." Ninety days from December 11, 2025 lands around March 11, 2026; the statement did not appear until July 7 - about four months past the deadline the order itself set.

What it says, once it arrives, is narrower than "state AI law loses." The FTC's theory is that a company deceives consumers if it "steers" an AI system's outputs toward objectives the user didn't ask for and wouldn't expect, without saying so. Its worked example is direct: a company, the statement says, "might suppress accuracy and interpose other objectives, such as so-called 'equity,'" to avoid liability under a law like Colorado's, and some states have gone further, "ultimately requiring American companies to embed 'ideological bias within [their AI] models.'" But the statement does not say complying with those laws is itself the violation. It says the undisclosed part is: a company "can... clearly and conspicuously disclose that its systems are designed to produce outputs that prioritize certain objectives over what users request and otherwise expect," and that disclosure - if adequate to actually shift what a reasonable consumer expects - is the FTC's own stated way out. As of this writing, the document remains a proposed statement; the public comment period closed July 31, 2026, and the FTC has not published a final version.

Preemption in name, disclosure in practice

Neither the executive order nor the FTC's statement says outright that this is how the confrontation with state AI law was always going to resolve. That reading is this Article's own, built from what the two documents do rather than what either claims about itself: an order framed around freeing AI companies from state-mandated output alterations produced, as its first concrete enforcement mechanism, a federal rule that leaves those state mandates standing and adds a disclosure requirement most companies complying with them do not currently meet. Nothing in either document blocks Colorado, or any other state, from keeping its law. The FTC's tool reaches only the silence around compliance, not the compliance itself.

The asymmetry this creates

That distinction cuts two ways, and neither favors the company that did nothing. An organization that already built out a compliance program for a state law like Colorado's - testing for the outcomes the law targets, adjusting outputs where the law requires it - was, until now, one audience: the state regulator. The FTC's proposed theory adds a second audience with a different question: did you tell the people using the product that its outputs are shaped this way? A compliance file built to satisfy a state agency's disclosure and documentation requirements was not necessarily built to satisfy a user-facing "clear and conspicuous" consumer disclosure standard - those are different documents, written for different readers, and the FTC's proposed theory treats the second as the one that matters to it.

An organization that instead waited for federal relief from state AI compliance costs gets the opposite of what it was waiting for. The instrument built to blunt state law arrived describing a new way to be found liable for how that same state-law compliance is communicated, not a reason state compliance can be skipped.

What a well-run AI operation does differently

The operational lesson isn't about Colorado specifically, or even about this one proposed statement surviving comment in its current form. It's about treating "why did we shape this output this way" as a decision that gets written down for a user to read, not just for an internal compliance file. Where an AI system's outputs are adjusted to satisfy a state law's requirements, a well-run operation now has a reason - independent of whether the FTC's proposed statement is finalized as written - to put that adjustment in front of the person seeing the output, in plain terms, at the point where they'd notice it mattered. The FTC's own proposed standard for what counts as adequate disclosure is unsettled; building the habit of writing the disclosure down where the user can find it is not.

Source note. This Article is drawn from Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence" (signed December 11, 2025; published in the Federal Register December 16, 2025), and the Federal Trade Commission's proposed "Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems" (published in the Federal Register July 7, 2026; public comment period closed July 31, 2026). Both are primary government instruments, read directly rather than through secondary summary.