Trump signs voluntary AI safety pledge as FTC circles

The CEOs of Google, Anthropic, Meta, OpenAI, xAI and Nvidia gathered at the White House and signed the "White House Agreement on Superhuman Artificial Intelligence" after lunch with President Trump. Trump called the move "tremendous self-regulation," but the document contains no binding obligations, no sanctions and no enforcement mechanism — only a list of recommendations the companies "should implement" voluntarily.

The declaration sets out four general provisions: first, establish strong internal controls to monitor model capabilities and ensure they do not "go rogue" or breach systems unintentionally. Second, empower an internal team to conduct that monitoring and fix failures when they occur. Third, cooperate with a qualified external party to perform independent assessments. Fourth, ensure the board has a committee that receives reports on all these efforts. There is no mention of public transparency, mandatory reporting to authorities, or capability thresholds that would trigger a pause.

The signing follows weeks of debate over AI regulation and coincides with a request the frontier labs filed for an antitrust exemption, which would let them coordinate safety work without fear of cartel suits. Antitrust experts rejected the request: if superhuman AI truly threatens human existence, as researchers warn, then it is anti-competitive by nature, because humans are a necessary condition for any market. In the background, the Federal Trade Commission has already said it intends to investigate several AI companies for suspected consumer-protection violations, quickly dashing any hope of a blanket "blessing" from the administration.

This is not the first voluntary pact the companies have signed. In early 2025 the United Kingdom and South Korea announced "Frontier AI Safety Commitments" that included promises of internal and external red-teaming for severe and novel threats, and for sharing safety information. That agreement also lacked teeth, and the companies did not respond to requests for comment then, just as they have not now.

Neil Chilson, former chief technologist at the FTC, noted that the pledge *could* serve as a basis for enforcement if a company breaches it materially, because a public promise that goes unfulfilled may count as a deceptive practice under the FTC Act. But the commission's standard remedy in such cases is an order barring the company from continuing to misrepresent — not structural reform and not significant fines. A recent example: the FTC settled with three companies that had promised business customers the ability to "listen" to consumers through smartphone and smart-TV microphones; the outcome was a ban on misrepresentation, not a product shutdown.

The bottom line: the new document is mainly a well-framed statement of intent. It lets companies show "safety effort" without changing their development pace, and lets the White House claim a regulatory win without passing legislation. Anyone looking for a real brake on more powerful models will have to wait for Congress, or for the next FTC lawsuit.