Artificial intelligence (AI) developers could face mounting legal risks if they modify their models’ outputs to comply with state AI laws without clearly informing users. According to a proposed Federal Trade Commission (FTC) policy statement, such undisclosed modifications could violate longstanding federal consumer protection principles.
The proposal, issued July 1 in accordance with President Donald Trump’s December executive order, argues that AI companies may violate Section 5 of the FTC Act if they quietly steer model outputs toward objectives other than those users reasonably expect, even when doing so to satisfy state regulatory requirements. A Sheppard Mullin analysis says the proposal places AI developers in an increasingly difficult position as a growing number of states adopt AI laws that may conflict with the federal government’s consumer protection approach.
The proposed policy statement reflects the Trump administration’s broader effort to establish a national AI policy while discouraging what it characterizes as a patchwork of state AI regulation. The FTC argues that consumers reasonably expect AI systems to provide the most accurate and faithful responses possible to their requests and that undisclosed manipulation of those outputs constitutes deception regardless of the company’s motivation.
According to the proposal, AI companies have consistently marketed their products as tools designed to solve users’ problems by delivering accurate, objective answers. Those explicit and implicit representations, the FTC contends, create consumer expectations protected by Section 5’s prohibition on deceptive acts and practices. If a developer secretly alters outputs to satisfy regulatory mandates or other objectives without adequately informing users, consumers may be paying for a product that does not perform as advertised.
The Sheppard Mullin analysis says the FTC has effectively transformed what many companies viewed as a regulatory compliance issue into a traditional consumer protection matter. Rather than focusing on whether state AI laws should require certain outputs, the Commission frames the issue as whether consumers are being misled about how AI systems actually operate.
The proposal singles out Colorado’s Artificial Intelligence Act as an example of the tensions facing developers. According to the FTC, both the original and revised versions of the law create incentives for AI companies to prioritize objectives such as avoiding discriminatory outcomes over delivering what users might otherwise consider the most accurate responses. The Commission argues that if developers make those tradeoffs without disclosure, they risk deceiving consumers.
Sheppard Mullin notes that Colorado is unlikely to remain an isolated case. With states continuing to introduce and enact AI legislation governing model outputs and algorithmic decision-making, companies operating nationwide may find themselves navigating inconsistent legal obligations. A developer that modifies outputs to reduce exposure under state anti-discrimination laws could face a federal deception claim, while a developer that declines to make those changes could face state enforcement or private litigation.
The FTC cites research indicating that users accept AI-generated responses without additional fact-checking more than 90% of the time. That level of reliance, the Commission argues, heightens the potential consumer harm if AI systems intentionally produce responses shaped by undisclosed objectives. The proposal distinguishes such intentional steering from so-called AI hallucinations, which it says generally arise from technical limitations rather than deliberate design choices.
The Commission stops short of prohibiting companies from modifying outputs, but emphasizes that any such practices must be accompanied by clear, conspicuous and persistent disclosures. Disclosures buried in terms of service or fine print are unlikely to satisfy Section 5, it said, because they would not meaningfully alter consumers’ reasonable expectations regarding how AI systems function.
Sheppard Mullin recommends several steps for AI developers as the proposal moves through the rulemaking process. Companies should audit both their model outputs and public representations to ensure marketing claims about accuracy, objectivity or truth-seeking align with actual system behavior. They should also review whether existing disclosures are sufficiently prominent to satisfy the FTC’s demanding standard and consider submitting comments before the agency’s Friday (July 31) deadline.
