FTC AI policy statement raises disclosure risk for developers

The proposed FTC AI policy statement issued July 1 would expose artificial intelligence developers to federal consumer-protection risk if they change model outputs to satisfy state rules without telling users clearly. The Federal Trade Commission cited research indicating users accept AI-generated answers without additional fact-checking more than 90% of the time, a level of reliance the agency says could magnify harm from undisclosed design choices.

The proposal, issued in accordance with President Donald Trump’s December executive order, frames the issue under Section 5 of the FTC Act. According to the FTC, companies may violate the law’s ban on deceptive acts and practices if they quietly direct AI systems toward goals other than the ones consumers reasonably expect.

A Sheppard Mullin analysis says the proposal puts AI developers between state-level compliance obligations and federal consumer-protection standards, as more states enact laws governing AI outputs and algorithmic decisions.

What does the FTC AI policy statement say?

The proposed policy statement says consumers generally expect AI tools to give accurate and faithful responses to their requests, based on how companies market those products. The FTC argues that explicit claims about accuracy, objectivity and problem-solving, along with implied representations, can create expectations protected by federal law.

Section 5 is the FTC’s main consumer-protection authority for policing deceptive business conduct. In this proposal, the agency treats hidden changes to AI outputs as a potential deception issue when users are led to believe the product is operating in a different way.

The Commission does not say AI companies are barred from modifying model outputs. It says those practices must be accompanied by disclosures that are clear, conspicuous and persistent, and that notices buried in terms of service or fine print are unlikely to change what consumers reasonably expect from an AI system.

Why state AI laws create a compliance conflict

The proposal points to Colorado’s Artificial Intelligence Act as an example of the tension. According to the FTC, the original and revised versions of the law may encourage developers to prioritize objectives such as avoiding discriminatory outcomes over producing what users might regard as the most accurate response.

Sheppard Mullin says that dynamic may not be limited to Colorado. Companies offering AI products nationwide could face inconsistent obligations if state laws push developers to alter outputs while the FTC treats undisclosed steering as potentially deceptive.

Under that scenario, a developer that adjusts model behavior to reduce exposure under state anti-discrimination requirements could face federal scrutiny if users are not told about the change. A developer that declines to make those changes could still face state enforcement or private litigation, according to the firm’s analysis.

The FTC distinguishes intentional steering from AI hallucinations. In the agency’s account, hallucinations typically arise from technical limitations, while undisclosed output steering reflects a deliberate design choice that can change what a user receives from a product sold as accurate or objective.

What AI developers are being told to review

Sheppard Mullin recommends that AI developers examine both model behavior and public statements about their systems as the proposal proceeds. The firm says companies should check whether marketing claims about accuracy, objectivity or truth-seeking match how the system actually operates.

The firm also says developers should assess whether their disclosures are prominent enough under the FTC’s proposed standard and consider submitting comments before the agency’s Friday, July 31 deadline.

This story draws on original reporting from PYMNTS.