US tech outlet WIRED, citing sources, reports the White House will expand an AI
safety-testing framework requiring federal safety tests of frontier models
before public release. The framework, announced this month but not yet public or
detailed, currently covers closed-source models from Anthropic and OpenAI;
officials expect it to include open-source models within months once they reach
‘frontier’ capability comparable to Anthropic’s Mythos and OpenAI’s GPT-5.6.