Microsoft, Google, and xAI give the U.S. government early access to unreleased AI models for security testing

Microsoft, Google DeepMind, and xAI agreed to give the U.S. government early access to unreleased AI models for national security testing. OpenAI and Anthropic were already participating in the same review process.

The program runs through the U.S. Center for AI Standards and Innovation at the National Institute of Standards and Technology. The reviews are focused on demonstrable risks, including whether advanced models could be used to launch cyberattacks, assist chemical or biological weapons development, or corrupt the data used to train U.S. models.

OpenAI is also working with the group to test GPT-5.5-Cyber, a model variant designed for defensive cybersecurity work. The expanded review path adds more frontier model developers to a pre-release testing process that had previously involved a smaller set of companies.

The arrangement moves federal AI oversight closer to actual model deployment. Instead of relying only on public commitments or post-release evaluation, the government is now reviewing some models before broad launch.

Sources: Reuters — https://www.reuters.com/legal/litigation/microsoft-xai-google-will-share-ai-models-with-us-govt-security-reviews-2026-05-05/ | CNN — https://www.cnn.com/2026/05/05/tech/microsoft-google-xai-government-test-ai-models