Commerce Department Expands Pre-Deployment AI Vetting to Google DeepMind, Microsoft, and xAI

The Commerce Department's Center for AI Standards and Innovation announced May 5 that it has signed pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI, bringing the total number of major frontier AI developers under voluntary federal review to five. The agreements authorize CAISI, a component of the National Institute of Standards and Technology, to examine the companies' models before public release. Through these arrangements, CAISI will conduct pre-deployment evaluations and targeted research to assess frontier AI capabilities and advance AI security. The three firms join Anthropic and OpenAI, which entered analogous agreements in August 2024 under the prior administration. ^[1][2]

CAISI operates within NIST, which sits under the Department of Commerce's nonregulatory mandate. The agency is authorized to establish voluntary agreements with private-sector AI developers and to lead unclassified evaluations of AI capabilities that may pose risks to national security, with a focus on demonstrable risks including cybersecurity, biosecurity, and chemical weapons. To conduct those evaluations, developers frequently provide CAISI with models that have reduced or removed safeguards, and evaluators from across government may participate through the TRAINS Taskforce, an interagency group focused on AI national security concerns. ^[3][4] The TRAINS Taskforce, originally established by NIST in November 2024, brings together partners from across the federal government to identify, measure, and manage emerging national security and public safety implications of AI. Its participating agencies include the Department of Defense, the Department of Energy and ten of its National Laboratories, the Department of Homeland Security, and the National Institutes of Health. ^[4] The agency itself was reconstituted in June 2025, when Commerce Secretary Howard Lutnick renamed and restructured the former U.S. AI Safety Institute as CAISI; as of May 2026, the TRAINS Taskforce includes participants from more than 10 agencies. ^[3][4]

The new agreements are not created from scratch. They build on previously announced partnerships, which have been renegotiated to reflect CAISI's directives from the Secretary of Commerce and America's AI Action Plan. ^[1][2] Commerce Secretary Lutnick has stated that CAISI will focus on demonstrable risks such as cybersecurity, biosecurity, and chemical weapons, and will also investigate malign foreign influence arising from the use of adversaries' AI systems. ^[3] The agreements support testing in classified environments and were drafted with flexibility to respond rapidly to continued AI advancements. ^[1] CAISI has, to date, conducted more than 40 evaluations, including on models that have not yet been publicly released. ^[5][2] CAISI Director Chris Fall said in a statement that "independent, rigorous measurement science is essential to understanding frontier AI and its national security implications." ^[1][5]

Microsoft separately confirmed it signed agreements with CAISI in the U.S. and with the AI Security Institute in the United Kingdom to advance AI testing, including collaborative work to test its frontier models, assess safeguards, and help mitigate national security and large-scale public safety risks. ^[6] Under the CAISI agreement, Microsoft and NIST will collaborate on improving methodologies for adversarial assessments, testing AI systems in ways that probe unexpected behaviors, misuse pathways, and failure modes. ^[6] All five agreements remain voluntary. No company is legally compelled to submit a model for review, and no public process exists for what happens if a CAISI finding flags a model as unsafe. ^[7] The practical incentive, however, is significant: the agreements support information-sharing, drive voluntary product improvements, and ensure a clear understanding in government of AI capabilities and the state of international AI competition. ^[1]

The CAISI announcement coincides with an accelerating White House review of whether to impose mandatory pre-deployment screening. The Trump administration is considering an executive order to ensure new AI models are secure before public release; National Economic Council Director Kevin Hassett compared the proposed approach to how the Food and Drug Administration evaluates drugs for safety, describing a process by which AI would be "released in the wild after they've been proven safe, just like an FDA drug." ^[5] The White House has also been weighing the creation of a new AI working group that would explore potential oversight procedures, including plans to vet models before they are released to the public. ^[2] An executive order requiring pre-deployment review of frontier AI models would likely increase the workload at NIST's Center for AI Standards and Innovation. ^[5] If mandatory vetting legislation advances, the five labs that have entered voluntary agreements would hold a procedural head start under any future compliance framework. ^[7]

The regulatory trajectory marks a substantive shift from the administration's initial posture. President Trump initially took an unregulated approach to AI in the first year of his second term, with goals of accelerating AI innovation, building domestic AI infrastructure, and advancing international AI diplomacy and security. ^[8] The White House AI Action Plan, released July 23, 2025, names NIST in a large number of recommended policy actions. ^[3] The expansion of CAISI's evaluation program to cover all five major U.S. frontier AI developers represents the most direct federal assertion of pre-market national security oversight authority over advanced AI systems to date, short of a statutory or executive mandate, and it positions CAISI as the institutional infrastructure for any such mandate should one materialize.

—