AI firms to meet the White House on safety testing

The White House has invited OpenAI, Anthropic, Meta, and Google to discuss a new voluntary framework for testing frontier AI systems before release. The meeting follows recent reports of agentrelated security breaches at OpenAI and Anthropic, which have renewed attention on model safety. According to the report, the framework would allow companies to share frontier models with the government up to 30 days before launch for cybersecurity testing. The discussion is expected to cover how frontier AI is defined, whether open models are included, and who would run the testing process. The proposal comes as regulators in the U.S. and Europe increase scrutiny of advanced AI systems. Supporters say the approach could help identify dangerous capabilities earlier, but the framework would remain optional and its details would stay classified.