The US government has announced agreements with Google DeepMind, Microsoft, and xAI to review early versions of their AI models prior to public release. This initiative, led by the Center for AI Standards and Innovation (CAISI) within the US Department of Commerce, aims to assess the capabilities of advanced AI technologies and their implications for national security. CAISI director Chris Fall emphasized the importance of rigorous measurement science in understanding frontier AI. The agreements focus on identifying national security risks related to cybersecurity, biosecurity, and chemical weapons. CAISI has previously completed over 40 evaluations of unreleased models, highlighting the common practice of developers sharing models with reduced safety measures to facilitate comprehensive risk assessments. As concerns grow over the potential dangers posed by powerful AI models, such as Anthropic's Mythos, these collaborations are seen as essential for ensuring public safety. Microsoft has also announced a similar agreement in the UK, reinforcing the need for collaborative efforts in AI testing for national security.