The FTC just handed marketers and AI vendors a loaded choice: break federal law or break state law. Pick one.
On July 1, the Commission issued its proposed policy statement on "accuracy suppression" in AI systems. The message is blunt. If you adjust AI outputs away from what they're designed to do, even to comply with state anti-discrimination laws, you're deceiving consumers under Section 5 of the FTC Act. And deception doesn't need a malicious motive to be illegal.
This is not a clarification. It's a trap.
The Paradox That Broke AI Governance
Here's what just happened. Colorado's AI Act (revised May 2026) requires companies to avoid outputs that cause disparate impact. That's a state law mandate. To obey it, you suppress AI accuracy. You tweak the model. You add guardrails. You trade raw performance for fairness.
Now the FTC shows up and says: "You altered the AI away from its intended purpose. That's a misrepresentation to consumers. Section 5 violation."
Your defense? "But Colorado law required it."
The FTC's response: State law compliance is not a defense. Federal law preempts it.
Marketers and financial services firms that spent the last 18 months building bias-mitigation frameworks just found out those frameworks might be illegal at the federal level. Unless they disclose every deviation, every adjustment, every compromise. On the consumer-facing side. Up front.
Why Disclosure Doesn't Actually Solve This
The FTC acknowledges you can avoid Section 5 violations with disclosures. But those disclosures have teeth.
They can't be buried in terms of service. They can't be a small gray footnote. The FTC's language is explicit: disclosures must "plainly dispel the notion" that your AI is designed for accuracy. The more your disclosure conflicts with what users reasonably expect, the more prominent and persistent it needs to be.
Imagine the landing page: "Our AI system is deliberately less accurate than it could be to comply with state fairness requirements." That's not driving conversions. That's confessing to a limitation consumers didn't know existed.
The FTC is essentially requiring you to advertise the compromise. Loudly.
The Real Cost: Operational Paralysis
This creates a cascade of problems for teams trying to build compliant AI products.
If you're a lender using AI for underwriting, you've been taught to watch for disparate impact. The FTC just told you that removing that bias might be illegal. You now need legal review for every output adjustment. You need actuaries to model the trade-offs. You need compliance to vet the disclosures.
The timeline to launch just doubled. The cost just tripled.
If you're a marketer using recommendation engines, you can't just suppress certain recommendations to avoid offending demographics. You need to document why you suppressed them. You need to prove the suppression was narrow and justified. You need to disclose it if it's material to consumer choice.
What's material? The FTC found that consumers accept AI recommendations without fact-checking 90% of the time. So almost everything is material now.
The Version Nobody's Talking About: Financial Services Gets Uglier
The CFPB in 2023 warned that financial institutions relying on AI chatbots for legally mandated disclosures could violate consumer law if the AI got it wrong. The FTC's new policy soft-pedals that. It says hallucinations alone aren't Section 5 violations.
But misrepresenting the likelihood of hallucinations still is.
So if your AI chatbot provides lending information and occasionally hallucinates, that's OK. But if you hide how often it hallucinates, or minimize the risk in your marketing, that's deception.
You're now liable for being accurate about inaccuracy.
What Happens in August
The comment period closes July 31. Expect financial services groups to file comments arguing that the FTC's position creates impossible compliance burdens. Expect state attorneys general to file comments defending their state laws. Expect the FTC to ignore both and finalize this anyway.
The Administration's executive order in December 2025 already signaled where this was headed. The FTC is moving enforcement-first, philosophy-later.
By September, if you're deploying AI in consumer-facing contexts, you'll need:
- A documented audit trail of every output adjustment
- Legal review of fairness vs. accuracy trade-offs
- Compliance sign-off on disclosures
- Clear metrics on what's "material" to your users
That's not a governance framework. That's bureaucracy masquerading as regulation.
The Uncomfortable Bottom Line
The FTC just made the cost of AI compliance higher than the cost of not using AI at all for most companies.
The vendors who survive this will be the ones big enough to afford the legal infrastructure. Everyone else will either strip their AI down to vanilla use cases, or wait to see who gets enforcement actions first.
Your state law said don't discriminate. Your federal law says don't lie about not discriminating. The gap between them is where real AI deployment goes to die.
Welcome to 2026.