Google, Microsoft and xAI agree to US government AI testing programme

Google, Microsoft, and xAI have agreed to let the US Commerce Department’s CAISI test their AI systems before public release.

The federal program will evaluate models for safety, security, and national security risks, including cybersecurity and biosecurity concerns.

This expands earlier testing efforts that already included OpenAI and Anthropic, widening government oversight of frontier AI.

The move signals a shift toward pre-release evaluation of major AI models in the US.