NIST’s CAISI Signs Agreements With Google DeepMind, Microsoft and xAI to Expand U.S. Frontier-AI Testing

The National Institute of Standards and Technology said Tuesday that its Center for AI Standards and Innovation, or CAISI, signed agreements with Google DeepMind, Microsoft and xAI to expand the U.S. government’s frontier-AI testing program, including evaluations before and after models are publicly released.

In a press release issued from Gaithersburg, Maryland, and Washington, NIST said the agreements are designed to support pre-deployment evaluations, post-deployment assessments and other collaborative research aimed at measuring frontier AI capabilities and advancing AI security. Google DeepMind is Alphabet’s AI research lab, Microsoft is a major cloud and AI platform company, and xAI is Elon Musk’s AI company.

The significance of the arrangement is not just that more major developers are joining the federal testing effort. NIST said CAISI will be able to assess models before public release, conduct testing in classified environments and, in some cases, examine versions with key protections dialed back. According to the release, developers frequently provide CAISI with models that have “reduced or removed safeguards” so national-security-relevant capabilities can be evaluated.

NIST also said the agreements allow evaluators from across the U.S. government to take part in testing. The agency said CAISI convenes the interagency TRAINS Taskforce — short for Testing Risks of AI for National Security — to provide feedback on evaluations. To date, NIST said, CAISI has completed more than 40 evaluations, including on state-of-the-art models that have not been released publicly.

The announcement further cements CAISI’s role as the government’s main hub for this work. NIST said Commerce Secretary Howard Lutnick designated CAISI as industry’s primary point of contact within the U.S. government for model testing, collaborative research and best-practice development. In a June 3, 2025, Commerce Department statement about remaking the institute, Lutnick said, “CAISI will evaluate and enhance U.S. innovation of these rapidly developing commercial AI systems while ensuring they remain secure to our national security standards.”

CAISI is the current name of what had been the U.S. AI Safety Institute. The Commerce Department announced the change in June 2025, recasting the organization around standards, innovation and security testing. The new agreements extend an existing federal approach rather than creating one from scratch: NIST had already announced similar partnerships with Anthropic and OpenAI on Aug. 29, 2024. NIST said those earlier arrangements have since been renegotiated to reflect CAISI directives and the U.S. AI Action Plan.

Chris Fall, CAISI’s director, said in Tuesday’s release that the broader set of agreements comes at an important moment for measuring fast-moving AI systems. “Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” Fall said. “These expanded industry collaborations help us scale our work in the public interest at a critical moment.”

Taken together, the new deals broaden the federal government’s formal testing network to three more leading AI developers while emphasizing a specific capability NIST says CAISI already has: evaluating frontier models before deployment, including in classified settings and sometimes with safeguards reduced or removed.