U.S. agency gains pre-release access to AI models from five major labs for national-security testing

The Center for AI Standards and Innovation (CAISI), part of the U.S. Department of Commerce, has signed agreements with Google Deepmind, Microsoft and xAI to conduct national — security evaluations of advanced AI systems before they reach the public. Reported May 5, 2026, the new deals expand CAISI’s pre-release testing partnerships beyond the two labs it previously worked with.

Under the arrangements, participating companies will supply pre-release model versions with reduced safety guardrails so CAISI can probe vulnerabilities and model behaviour under realistic threat scenarios. The agency will run evaluations in classified environments; CAISI says it has already carried out more than 40 tests, including some on unreleased models. Those controlled assessments are intended to reveal exploitable flaws that might not appear in public releases.

The expansion builds on earlier agreements CAISI struck with Anthropic and OpenAI covering joint safety assessments and collaborative research into risk mitigation. Officials framed the effort against a backdrop of rapidly improving models that can find and exploit security weaknesses and an intensifying technology competition with China. CAISI Director Chris Fall said, “Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” and added that broader pre-release access and classified testing should deepen joint safety work and support research to mitigate AI-driven national security risks.