Washington Gains Access to AI from Google, Microsoft, and xAI

Strategic Access for National Security

The United States government has gained early access to artificial intelligence models developed by five major laboratories as part of national security testing. Among the entities involved are Google Deepmind, Microsoft, and xAI. These agreements were orchestrated by the Center for AI Standards and Innovation (CAISI), a branch of the U.S. Department of Commerce, with the aim of assessing potential risks associated with these technologies before they are made available to the public.

Rigorous and Independent Assessments

Chris Fall, director of CAISI, emphasized the importance of independent measurement science to understand the implications of advanced AI technologies on national security. To date, the agency has already conducted over 40 assessments, some of which focus on models that have not yet been released. The involved AI laboratories provide versions of their models with reduced security measures to facilitate thorough testing.

Strengthening Existing Agreements

These new agreements build upon those already established with Anthropic and OpenAI. They allow for testing in secure and classified environments. The initial agreements already included joint security assessments and research on risk mitigation. This expansion comes as the capabilities of AI models advance rapidly, particularly in detecting and exploiting security vulnerabilities, and as technological competition with China intensifies.