Google, Microsoft, and xAI Share Unreleased Models With U.S. Government for Security Testing
Google, Microsoft, and xAI have agreed to provide the U.S. government with unreleased versions of their AI models for advance testing. The Center for AI Standards and Innovation (CAISI), part of the Department of Commerce, will conduct the evaluations before these systems reach the public.
CAISI will assess frontier AI models and LLMs for national security risks including cybersecurity threats, biosecurity vulnerabilities, and potential use in chemical weapons development. The arrangement gives federal agencies early access to commercial systems before their public release.
The agreement expands on earlier partnerships. OpenAI and Anthropic made similar commitments to the Biden administration roughly two years ago. CAISI has since conducted dozens of evaluations of advanced models, including systems not yet available to the public.
Chris Fall, CAISI director, said independent and technically rigorous evaluation methods are necessary to understand how frontier AI affects national security. He added that the expanded industry collaboration allows the institute to conduct security reviews faster and at greater scale as AI technology develops rapidly.
Government Shifts Toward Active Oversight
The announcement reflects a notable policy change. The Trump administration has taken a cautious approach to AI regulation, prioritizing innovation speed to maintain technological advantage over China. Yet concerns about AI risks are growing within Washington.
The recent public release of Anthropic's Claude Mythos model reignited debate about the pace of AI system development. The New York Times reported this week that the Trump administration is drafting a potential executive order on AI governance. Under the proposal, technology companies and government agencies would jointly establish a formal review process for new models.
This marks a shift from the White House's previous hands-off stance. Trump stated last year that artificial intelligence cannot be slowed by political measures or excessive regulation. The administration now appears to be moving toward a more active oversight role, driven partly by public concerns about cybersecurity, job losses, misinformation, and mental health impacts.
For AI for Government professionals, this development signals that federal AI policy is entering a new phase-one where pre-deployment security reviews and industry-government collaboration are becoming standard practice.
Your membership also unlocks:
