The Dilemma
Your AI system's outputs might trigger liability under state law. You can modify the model to reduce this risk, but doing so may mean the system won't always meet user expectations. Should you prioritize state compliance or user expectations?
This isn't just theoretical. The FTC's proposed policy statement from July 1, 2026, frames this as a federal deception issue. If you alter AI outputs to comply with state requirements and those changes conflict with user expectations, you might violate Section 5 of the FTC Act. The comment period closes July 31, 2026, but the underlying tension will persist.
Prioritizing User Expectations
The FTC's stance is clear: your AI system's promises create enforceable expectations. If you market your model as accurate and objective, users expect outputs that align with those claims. Introducing undisclosed objectives means you're not just changing behavior; you're misrepresenting the system.
This theory is based on established deception principles. Misstatements matter, but so do omissions and implied representations. If your marketing emphasizes accuracy and your terms of service don't clarify that you're steering outputs to avoid state liability, you've created a gap between promise and performance. The FTC differentiates this from AI hallucinations: technological limitations are one thing, intentional steering toward undisclosed objectives is another.
Consider the disclosure burden. The FTC states that clear and conspicuous disclosures can mitigate liability, but they can't be hidden in fine print. If you're modifying outputs to comply with Colorado's Artificial Intelligence Act, which imposes liability for discriminatory outcomes, you need to inform users that compliance objectives shape what they see. That disclosure must be prominent enough to alter their expectations before they rely on the output.
The preemption argument supports this position. The FTC suggests that state laws requiring conduct it views as deceptive may conflict with federal consumer protection law. Executive Order 14365 directed the FTC to address how Section 5 applies when developers alter outputs in response to state requirements. This signals a federal interest in limiting what the administration sees as a patchwork of state AI rules that could force companies into deceptive practices.
Prioritizing State Compliance
State AI laws aren't optional. Colorado's statute creates real liability exposure for discriminatory outcomes. You can't ignore it because the FTC prefers a different approach. Until a court invalidates the state law or federal legislation preempts it, you're subject to both regimes.
The compliance approach is straightforward: modify outputs to reduce state liability, then disclose that modification. The FTC acknowledges that disclosure can address deception concerns. If you explain that your system incorporates fairness constraints required by state law, you've given users the information they need to adjust their expectations. You're not hiding the steering; you're documenting it.
This approach reflects legitimate business judgment about competing risks. A state enforcement action carries immediate consequences: penalties, injunctions, reputational harm. The FTC's proposed policy statement, by contrast, is just that: proposed. It hasn't been finalized, and even after finalization, enforcement would require the FTC to prove that your disclosures were inadequate and that users were actually misled. That's a higher bar than state liability for discriminatory outcomes, which may hinge on statistical analysis of results rather than intent or disclosure.
You can also argue that users' reasonable expectations should include awareness of legal constraints. No one expects an AI system to break the law to serve them. If state law prohibits certain outputs or requires certain modifications, users should reasonably expect compliance with that law as part of the system's design. The FTC's focus on "undisclosed objectives" suggests that disclosure solves the problem, not that state compliance itself is impermissible.
Where Practitioners Actually Land
Most regulatory affairs teams treat this as a disclosure design problem, not a binary choice. You're not abandoning state compliance or ignoring user expectations. You're figuring out how to satisfy both through transparency.
This means mapping where state requirements change outputs, then building disclosure mechanisms that surface those changes at the point of interaction. If Colorado law drives you to adjust hiring recommendation outputs, your interface needs to explain that adjustment when users access those features. Generic terms of service language won't suffice. You need context-specific disclosure that explains what the system is doing and why.
You're also reviewing your marketing claims. If you've positioned your AI as "unbiased" or "purely objective," you've created a representation that conflicts with any intentional steering, including state-mandated steering. Revising those claims to acknowledge legal compliance constraints gives you more room to operate without triggering FTC deception theories.
The preemption question remains unresolved. Until courts weigh in or Congress acts, you're planning for both scenarios: one where state laws stand and you need robust compliance programs, and one where federal principles limit state requirements and you need documentation showing you prioritized user expectations.
Our Take
Prioritize transparency over choosing sides. The FTC's proposed statement treats disclosure as the mechanism that reconciles state compliance with federal consumer protection principles. That's your path forward.
You can't ignore state law. You also can't bury output modifications in legal boilerplate and claim you've disclosed them. The practical standard is whether a user would understand, before relying on an output, that the system is pursuing objectives beyond accuracy and usefulness. If state law requires fairness constraints that affect hiring recommendations, say so in the interface. If compliance with state requirements means certain queries won't return the most statistically likely answer, explain that tradeoff.
This approach acknowledges the real tension: state laws may require conduct that conflicts with user expectations shaped by your marketing. But that conflict doesn't force you to choose one regime over the other. It forces you to be explicit about what your system does and why. The FTC's focus on "undisclosed objectives" creates a compliance pathway through disclosure design, not through abandoning state law obligations.
The comment period through July 31, 2026, matters. If your organization has evidence that the FTC's approach creates unworkable conflicts with specific state requirements, submit it. If you've developed disclosure frameworks that satisfy both user expectations and state compliance, document them. The final policy statement will shape how Section 5 applies to AI systems for years, and your input can influence whether that framework is workable or creates impossible tradeoffs.
