White House may extend AI safety framework to open-weight models

The White House is reportedly considering expanding its voluntary AI model-evaluation framework, which currently screens only closed frontier models for cybersecurity risk, to also cover open-weight models. The existing framework gives labs up to 30 days to submit eligible models for review before public release.

Why it matters: Excluding open models from the current framework was widely read as a nod to domestic open-source competitiveness against Chinese labs; broadening it would mark a shift toward treating open-weight releases as a comparable risk category to closed frontier models. Worth watching alongside the Anthropic IPO's China-risk scrutiny and the reasoning-trace-extraction stories already covered.