Already a subscriber? Make sure to log into your account before viewing this content. You can access your account by hitting the “login” button on the top right corner. Still unable to see the content after signing in? Make sure your card on file is up-to-date.

The Trump administration has briefed the biggest AI companies on a finalized framework letting the federal government test their most advanced models before public release.

Getting into it: Representatives from OpenAI, Anthropic, Google, Meta, Microsoft, and Nvidia attended the White House meeting. Under the framework, which came out of a June executive order, companies can hand over a new model on their own initiative anytime in the month leading up to launch. Anything submitted gets run through a classified benchmarking system built to score cyber capability, and the results go out to federal agencies and a set of trusted corporate partners.

Only closed, proprietary US models demonstrating state-of-the-art hacking capabilities are covered. Open-weight models, the kind users download and modify themselves, are exempt. That likely captures Anthropic, OpenAI, and Google while leaving out Meta and Nvidia, at least for now. The Chinese open-weight models from Alibaba, DeepSeek, and Moonshot AI that worry administration officials most fall outside it too, since Beijing is unlikely to submit anything for American review.

The testing criteria will not be made public, and the government will not name the outside institutions cleared to see advanced models early.

The White House pitched the framework as a security win, saying, “The voluntary framework advances the Trump administration’s America First cybersecurity strategy by strengthening our national security and cementing American A.I. dominance.”

This all comes after OpenAI and Anthropic both disclosed that their models slipped restrictions in testing environments and broke into outside services. OpenAI said Tuesday it found additional smaller attacks, including one during testing by Britain’s AI Security Institute, which reported that Anthropic’s models had gone after outside targets too.

The House Homeland Security Committee has asked Sam Altman to brief lawmakers on the breach at AI development platform Hugging Face.