The FTC’s proposed AI accuracy policy would not automatically invalidate state chatbot laws. Instead, it would create a federal enforcement position—and potentially support court challenges—against state requirements that allegedly force companies to steer AI outputs away from users’ reasonable expectations of truthful or accurate answers.
Main effect: pressure on output-steering requirements
The FTC argues that secretly modifying an AI system’s outputs to pursue an undisclosed objective—such as an ideological, political, or certain “equity” objective—could be deceptive under Section 5 of the FTC Act. Compliance with a state law would not, in the FTC’s view, provide a defense to federal liability. ftc
The agency’s strongest target appears to be state requirements that could cause developers to suppress or alter otherwise accurate outputs to avoid disparate-impact or discrimination liability. Colorado’s AI framework is identified as an example. Such provisions could face arguments that they are impliedly preempted where compliance with state law conflicts with federal consumer-protection objectives. ftc
That would not mean those state laws are already invalid. A court would still have to determine, among other things, whether the state requirement actually requires inaccurate or misleading output, whether the federal and state regimes can be followed simultaneously, and whether the state law regulates output content or instead regulates downstream discriminatory use.
Most chatbot disclosure and safety laws would likely remain viable
State laws requiring a chatbot to:
- disclose that it is an AI system;
- avoid falsely claiming to be human;
- provide protections for minors;
- use self-harm or crisis protocols;
- restrict certain harmful content; or
- provide safety notices and records
would not ordinarily require the chatbot to give false answers. Those laws regulate transparency, safety, or product conduct rather than hidden manipulation of factual outputs. They would therefore be less directly affected.
Similarly, the FTC policy appears to distinguish intentional output steering from ordinary technical errors such as hallucinations. An accidental error would not automatically constitute a Section 5 violation, although false marketing claims about a system’s accuracy, reliability, or neutrality could create separate risk.
Disclosure could provide a way to comply with both regimes
The proposed policy indicates that companies may reduce deception risk by clearly and prominently telling users when a system prioritizes objectives such as safety, fairness, or legal compliance over unrestricted accuracy or the user’s requested objective. A disclosure buried in terms of service would likely be insufficient; it would need to be conspicuous and persistent enough to change users’ expectations. ftc
This could encourage states to draft chatbot laws around disclosure, documentation, testing, and risk management rather than prescribing particular answers or requiring undisclosed changes to model behavior.
Practical consequences
Until courts or Congress resolve the issue, companies could face competing risks:
- follow a state output-related requirement and risk an FTC deception claim;
- follow the FTC’s position and risk state enforcement;
- create different model behavior for different jurisdictions;
- use prominent disclosures; or
- challenge a state requirement as preempted.
The policy is currently a proposed policy statement, not a statute, final regulation, or automatic federal override of state law. Public comments are due July 31, 2026, and its final language or enforcement approach could change. ftc
Bottom line: state chatbot laws focused on disclosure, minors, safety, impersonation, and privacy would probably be largely unaffected. State provisions that allegedly require hidden suppression or distortion of accurate answers—particularly in the name of fairness or disparate-impact prevention—would face substantially greater federal preemption and FTC-enforcement pressure.
