US Government's CAISI to Pre-Test Google, Microsoft and xAI Frontier AI Models Before Public Release

The Center for AI Standards and Innovation (CAISI), housed within NIST at the U.S. Department of Commerce, announced on May 5, 2026 that it has signed pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI. The deals require the companies to provide access to frontier AI models before public release for government safety and capability assessments.

Scope of Evaluations

CAISI will conduct "pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security," including testing for cybersecurity, biosecurity, and chemical weapons risks. Evaluations occur in classified environments. Developers sometimes provide versions of models with reduced or removed safeguards to allow more thorough national security assessments.

Building on Prior Agreements

CAISI previously signed similar agreements with OpenAI and Anthropic in 2024 and has completed more than 40 evaluations to date, including models that remain unreleased publicly. The new agreements extend this framework to cover every major frontier AI developer operating in the United States.

Significance

The development is notable given that the Trump administration came into office opposing AI oversight. The willingness to expand pre-release model testing suggests that national security considerations create bipartisan support for AI governance even when broader regulatory frameworks remain contested. Coverage: CNBC, HPCwire.