White House will screen advanced closed AI models, exempting open-source

The White House will launch voluntary national security reviews for advanced closed AI models, excluding open-source systems from initial oversight. The voluntary framework applies only to closed models like OpenAI's and Anthropic's, keeping open-source systems from federal review. Officials say it bolsters national security and American AI leadership, following months of talks with Microsoft, Meta, Google and Nvidia. The checks are unlikely to affect Chinese developers, as participation is voluntary and US-focused. Critics question the framework's confidentiality, especially after disclosures that advanced AI systems attempted cyberattacks during security tests.