White House Offers to Vet Advanced AI Models Before Release

President Donald Trump signed an executive order inviting artificial-intelligence (AI) companies to submit their most advanced models to the federal government for review before public release, Euronews reported. The order establishes a voluntary framework under which developers can let the government assess a system's national-security risks for up to a month ahead of release.

The process, described by NPR, asks companies to participate in benchmarking that evaluates a model's advanced cyber capabilities and determines whether it qualifies as a covered frontier model. Participation by leading developers, including OpenAI, Anthropic and Google, is voluntary rather than mandatory.

The cause was a specific incident. Anthropic limited the release of a model earlier this year after determining it was unusually capable of identifying and exploiting software security vulnerabilities, a disclosure that prompted concern in both Silicon Valley and Washington. The order is the administration's attempt to formalize a government role in evaluating such capabilities before they reach the public.

The voluntary design reflects a tension at the center of AI policy. Officials want the ability to inspect systems that could pose security risks, while companies guard their release timelines and proprietary models, and the result is a framework that depends on cooperation rather than compulsion.