US government now has pre-release access to AI models from five major labs for national security testing

The US Department of Commerce now has pre-release access to AI models from five major labs for national security testing. Anthropic, OpenAI, Google DeepMind, Microsoft, and xAI have signed agreements with the Center for AI Standards and Innovation, providing the government with versions of their models featuring reduced safety guardrails for evaluation in classified environments.

This expansion builds on earlier agreements with Anthropic and OpenAI. The program addresses two core government priorities: identifying AI vulnerabilities before public release and maintaining technological parity with China as competition intensifies. Testing occurs in isolated, secure facilities where researchers can probe models for weaknesses without triggering standard safety restrictions.

The reduced guardrails matter because they allow testers to explore edge cases and potential misuse scenarios that normal safety filters would block. Government researchers can assess how these systems respond to requests involving weapons development, cyberattacks, biological threats, and other national security concerns. This testing informs both defensive strategies and policy decisions.

The arrangement reflects a delicate balance. The companies retain control of their models and intellectual property while providing government access to frontier capabilities. The classified testing prevents sensitive findings from becoming public. Yet it also normalizes government access to AI systems and raises questions about oversight mechanisms and what happens to the data collected during testing.

The timing signals escalating stakes in AI development. US officials increasingly view advanced AI as strategic infrastructure comparable to nuclear weapons or semiconductors. By integrating government security testing into development cycles, the Department of Commerce aims to catch dangerous capabilities before deployment while ensuring American labs stay ahead technologically.

China's AI ambitions and the rapid pace of model releases from competing labs create pressure to move quickly. The program attempts to split the difference between safety and speed, testing systems early while they remain mostly theoretical rather than widely deployed.

KEY INSIGHT: The government is building real-time security vetting into AI development rather than reacting after public release