The Trump administration has finalized voluntary cybersecurity tests that measure the hacking capabilities of the most advanced American AI models, a White House official said Monday, as Meta, OpenAI, Anthropic and Google were invited to a closed-door meeting to discuss them.
The framework stems from a June 2 executive order that lets the government review powerful new models for national security risks for up to 30 days before launch. Officials did not disclose what the document contains, how results would be reported or when companies would begin using it. Google, OpenAI and Anthropic reviewed a draft and submitted edits in late July, Politico reported.
OpenAI asked the administration to put Commerce Department AI safety specialists at the center of any testing, and pointed to China’s more centralized approach to AI.
The meeting follows disclosures that Anthropic and OpenAI models breached other companies’ systems during tests, stirring lawmakers’ concerns about AI-driven cyberattacks. Five Senate Democrats, Kirsten Gillibrand, Chris Coons, Mark Kelly, Adam Schiff and Mark Warner, pressed the administration for details, warning that its “unpredictable approach” pushes customers toward cheaper Chinese models.
