Google DeepMind, Microsoft, and xAI have signed agreements giving the US government early access to unreleased frontier AI models for national-security testing, expanding Washington's ability to assess advanced commercial systems before they reach the public.
The Center for AI Standards and Innovation, or CAISI, under the US Commerce Department's National Institute of Standards and Technology, said on May 5 that the agreements will support pre-deployment evaluations and targeted research into AI capabilities and security risks.
US expands model testing
CAISI said the agreements build on earlier partnerships with OpenAI and Anthropic and have been updated to reflect Commerce Department directives and America's AI Action Plan.
The center has completed more than 40 evaluations, including assessments of advanced models that have not been released. CAISI said it has been designated as the primary US government contact for industry collaboration on commercial AI system testing, research, and best-practice development.
The agreements allow CAISI to evaluate models before deployment, conduct post-deployment assessments, and pursue related research. The agency said the arrangements are intended to support information-sharing, encourage voluntary product improvements, and give the government a clearer view of frontier AI capabilities and international competition.
Safeguards may be reduced for testing
To assess national-security risks, developers may provide CAISI with models that have reduced or removed safeguards, the agency said. Government evaluators may participate through the CAISI-convened TRAINS Taskforce, an interagency group focused on AI national-security concerns.
The agreements also support testing in classified environments and are designed to give the government flexibility as AI systems continue to advance.
"Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications," CAISI Director Chris Fall said in the announcement. "These expanded industry collaborations help us scale our work in the public interest at a critical moment."
Reuters reported that Microsoft will work with US government scientists to test AI systems for unexpected behaviors and develop shared datasets and workflows for evaluating its models. The agreements come as US officials grow more concerned about national-security risks from powerful AI systems, including their potential use in cyberattacks or military applications.
Broader review remains under consideration
The Wall Street Journal reported separately that the Trump administration is considering a broader cybersecurity-focused review process for AI tools deemed to pose risks. The possible executive order could formalize a government oversight group to create standards for the most powerful AI models, according to the report.
That broader review process has not been finalized. The confirmed agreements announced by CAISI are focused on industry collaboration, pre-deployment evaluations, and security research involving Google DeepMind, Microsoft, and xAI.
Article edited by Jack Wu
