President Trump announced a voluntary AI safety accord that relies on large technology firms to police themselves. Dozens of companies, including Anthropic, OpenAI, SpaceX AI, Nvidia, Meta, and Alphabet/Google, signed on to conduct safety tests, meet regularly, and develop common standards. The agreement was posted on Trump’s Truth Social account, but detailed procedures have not been released.
The pact carries no new legal obligations; Trump described it as ‘morally binding’ despite lacking statutory force. Several signatories such as Anthropic, Google, and OpenAI had already pledged external audits before the deal. Some firms also agreed to participate in confidential government safety testing, a practice reported by The Information. Trump suggested future codification of the recommendations, but for now the administration emphasizes industry self-regulation.
Details of the new safety audits remain sparse. Bloomberg noted that the selection of third-party reviewers appears politicized, leaving firms uncertain about qualifying auditors. The Center for AI Standards and Innovation reported that several prospective reviewers have already met with the U.S. agency overseeing voluntary pre-deployment safety tests. This limited transparency has prompted questions about how frontier models will be evaluated.
Critics view the accord as a political gesture ahead of the mid-term elections. Bloomberg cited a Quinnipiac University poll showing more than seven-in-ten voters favor candidates who support stricter AI guardrails. Senator Bernie Sanders accused Trump of prioritizing industry interests, highlighting a state dinner photo that featured tech CEOs alongside Chinese President Xi Jinping and Melania Trump. The administration’s rhetoric framed AI leaders as the world’s most brilliant people.
Industry insiders remain skeptical that self-policing will resolve ongoing safety challenges. OpenAI, which has faced several high-profile incidents, relies on a Safety and Security Committee chaired by computer-science professor Zico Kolter. Kolter told NBC News that governance structures must evolve to meet emerging risks, emphasizing that the committee’s effectiveness depends on developers reporting concerns. OpenAI has not disclosed recent changes to its early-stage testing, and the company declined to comment when queried.
Regulators are pursuing parallel oversight. The Federal Trade Commission announced a broad investigation into consumer harms from rogue AI agents at firms such as Anthropic and OpenAI, according to The New York Times. State attorneys general, including California’s, have pressed OpenAI for greater transparency, with Encode’s general counsel Nathan Calvin signing an open letter urging the Safety and Security Committee to disclose its effectiveness. These actions suggest that governmental scrutiny may intensify despite the voluntary nature of Trump’s accord.