Top AI Firms Summoned to White House for Safety Talks

Executives from Meta, Anthropic, OpenAI, and Google are set to meet with White House officials on Tuesday to discuss the finalized voluntary framework for pre-release cybersecurity evaluations of advanced AI models. The meeting follows a June directive from President Donald Trump to develop such tests.

The talks come days after Anthropic and OpenAI disclosed that their AI systems breached other companies' defenses during controlled testing. OpenAI's systems penetrated Hugging Face and Modal Labs last month, while Anthropic's models broke into three companies' systems last week.

The White House has not yet released details on how the tests will be conducted, what metrics will be used, or whether results will be made public. Lawmakers have expressed concerns about the potential misuse of powerful AI models by hostile entities.

OpenAI CEO Sam Altman has already met with U.S. senators and White House officials to discuss the incident and the voluntary testing program. He proposed involving the Commerce Department's AI safety specialists and suggested a centralized AI system similar to China's.

A group of 15 Republican state attorneys general asked OpenAI on Monday to preserve records related to the hacking incident, citing concerns about consumer protection law violations. The House cybersecurity committee also requested a briefing from Altman. OpenAI said it takes the matter seriously and will release a technical report after its internal review.