The FTC Wants a Say in Whether Your AI Is Telling the Truth — Here's What That Actually Means

On July 1, 2026, the Federal Trade Commission opened a public comment window on a proposed policy statement that could reshape how AI companies talk about — and quietly modify — the outputs of their models. If you use AI in your business, this is one of the most important regulatory signals of the year, and it landed with almost no fanfare.

Here is the plain-English breakdown of what the FTC is actually proposing, why it matters for businesses buying AI, and what the fight with state laws like Colorado's Artificial Intelligence Act really means.

What the FTC Is Actually Proposing

The proposed policy statement invokes Section 5 of the FTC Act, which prohibits "unfair or deceptive" acts or practices in commerce. The Commission argues that when an AI company:

• Advertises its system as objective, accurate, or generally suitable for a task, and
• Silently manipulates model outputs to advance undisclosed ideological objectives,

…that gap between the marketing and the reality can amount to deception. In other words, the FTC is telling AI vendors: if you say your model is neutral and it isn't, that's not just a values debate — it's potentially a Section 5 violation.

One line in the statement matters more to buyers than any other. The Commission's position is that an AI company pursuing undisclosed ideological objectives may be deceiving consumers "whether of its own volition or in response to a state law requirement." Read that again: complying with a state law is not, in the FTC's view, a defense to a federal deception claim. That is what puts vendors in a genuine bind, and it is why the Colorado discussion below matters.

The statement also sets a high bar for what counts as adequate disclosure. It has to be prominent and persistent — not a line buried in terms of service or in technical documentation. If your vendor's answer to "do you steer outputs?" is a clause in a document nobody reads, that is unlikely to cure the problem.

Announcing the proposal, FTC Chairman Andrew N. Ferguson said the Commission "wants to hear from businesses and consumers about their experiences and concerns regarding the subversion of AI systems for ideological ends" (FTC press release, July 1, 2026). The statement was issued pursuant to Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence" (Dec. 11, 2025), which directs the Commission to clarify how Section 5 applies to AI models — including where state laws require companies to alter accurate model outputs.

The Colorado Preemption Fight

The proposed statement singles out Colorado's Artificial Intelligence Act as an example of a state law that "appears to coerce companies into altering the output of their AI models to comply with and advance the state's ideological objectives" (proposed policy statement, PDF). Colorado has since revised that law — S.B. 26-189, § 6-1-1707, enacted May 14, 2026 — and the revised version explicitly provides that AI companies can be held liable for discriminatory outcomes.

Here it is worth being precise about what the FTC did and did not say, because the distinction gets flattened in most coverage. The statement recites a general principle of law: "State law is impliedly preempted to the extent it conflicts with a Federal regulatory scheme." That is black-letter preemption doctrine, not a finding. The FTC did not conclude that Colorado's Artificial Intelligence Act is preempted. It raised the conflict question, named Colorado as the example, and positioned itself to argue the point later.

That is still a meaningful signal — agencies do not name a specific state statute by accident. But it is a signal, not a holding, and no business should treat Colorado's obligations as suspended. Expect a federal-versus-state fight over AI output rules; do not expect it to be resolved soon. Businesses operating across states — especially in Colorado, Illinois, California, and New York — should expect their AI vendors to push back harder on state-specific compliance demands while the question is open.

Why This Matters for Businesses Using AI

This isn't just AI-company drama. If you deploy AI in customer-facing workflows — chatbots, research summaries, marketing copy, financial or legal drafting, hiring tools, real estate valuations — you are the one presenting those outputs to your customers. If the underlying model has undisclosed filters that materially affect accuracy in your domain, the deception risk starts moving downstream to you.

Three practical implications:

1. Vendor due diligence just got harder. "Does the model work well?" is no longer enough. You need to ask: What content, viewpoints, or outputs are being suppressed or reshaped, and is that disclosed to end users? Get it in writing.
2. Disclosure language in your product needs a review. If your AI-powered feature is marketed as "unbiased," "objective," "accurate," or "based on the facts," and your vendor is quietly steering outputs, you inherit the disclosure gap. Update your marketing copy and terms of use.
3. Multi-vendor strategies get more valuable. If a single provider's undisclosed filters could distort your workflow, you want the ability to swap models — Claude, Gemini, GPT, open-source — without rebuilding your product. Design for portability.

What This Is Not

This policy statement is not a rule, not a fine, and not a lawsuit. It is a proposal the FTC announced on July 1, 2026 and published in the Federal Register on July 7, 2026 (91 Fed. Reg. 41638), with a comment window that ran through July 31, 2026. Once finalized, it becomes enforcement guidance — a signal about how the Commission will interpret Section 5 in future cases. It does not itself change the law.

But policy statements are how the FTC telegraphs which cases it will bring. Vendors, plaintiffs' lawyers, and state AGs all read them carefully — and they shape settlement negotiations long before an actual case reaches court.

What to Do About It

• If you sell an AI-assisted product: audit your marketing claims about accuracy, objectivity, and neutrality. Fix anything that overstates what your vendor actually delivers.
• If you buy AI from a major vendor: add a due-diligence question about output filtering, content policies, and disclosure practices to your next renewal. Ask specifically whether disclosure is prominent and persistent, since that is the standard the FTC proposed — a clause in the terms of service will not carry the weight. Vendors will get used to being asked.
• If you operate in multiple states: flag Colorado, California, Illinois and New York AI requirements for legal review. Keep complying with them — the FTC raised a preemption question, it did not answer one — but understand that your vendors may argue the other way.
• If you want to see how the industry is positioning: the comment window ran through July 31, 2026, and filings at Regulations.gov (docket FTC-2026-0859) are part of the public record. Reading what vendors and state AGs filed is the cheapest available preview of the fight.

Bottom Line

The FTC is putting AI vendors on notice that "objective" and "accurate" are not marketing adjectives anymore — they are claims the Commission is willing to test. For businesses building on top of those models, the takeaway is simple: understand what your AI is actually doing under the hood, disclose it honestly to your customers, and design your workflows so no single vendor's editorial decisions can quietly reshape your product.

Primary sources

• Federal Trade Commission, "FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy" — press release, July 1, 2026.
• Federal Trade Commission, Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems (PDF), Matter No. P264200.
• 91 Fed. Reg. 41638 (July 7, 2026), document 2026-13628.
• Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," 90 Fed. Reg. 58,499 (Dec. 11, 2025).
• Regulations.gov docket FTC-2026-0859 — public comments.
• Colorado S.B. 26-189, § 6-1-1707 (enacted May 14, 2026), revising the Colorado Artificial Intelligence Act.

Published July 19, 2026. Updated and fact-checked against the primary sources above on July 24, 2026.

Correction (July 24, 2026): An earlier version of this article stated that the FTC's position is that Colorado's Artificial Intelligence Act is "impliedly preempted to the extent it conflicts with a federal regulatory scheme." That quotation is a general statement of preemption doctrine appearing in the policy statement, not an FTC conclusion about Colorado's law. The section has been rewritten to reflect the distinction. This version also notes Colorado's May 2026 revision of the Act, adds the Commission's position that state-law compliance is not a defense to a federal deception claim, and removes quotation marks previously placed around a paraphrase of Executive Order 14365.