U.S. Government Secures Early Access to Frontier AI Models from Google, Microsoft, and xAI

On May 5 2026 CAISI, within the Department of Commerce’s NIST, announced agreements with Google DeepMind, Microsoft, and xAI to provide the U.S. government early access to AI models for pre-deployment testing, research, and post-deployment assessment.

Google DeepMind, Microsoft and xAI Sign Agreements for US National Security AI Testing

The pacts extend 2024 OpenAI and Anthropic deals, renegotiated under Trump administration’s AI Action Plan, after Anthropic’s Mythos model—described as “far ahead” in detecting vulnerabilities—was cited by officials as raising hacking-risk concerns.

Google, Microsoft, and xAI agree to let US government test AI models before public release — OpenAI and Anthropic also on board after renegotiating deals with Washington

CAISI, led by Director Chris Fall, oversees testing. Microsoft’s chief responsible AI officer, Natasha Crampton, highlighted CAISI’s technical and national-security expertise. Google DeepMind and xAI declined comment. The Defense Department stresses augmenting warfighter decision-making. Anthropic remains in a Pentagon-guardrail dispute.

CNN

CAISI reports over 40 AI model evaluations, including unreleased state-of-the-art systems. The agreements extend testing to the three signatories. Microsoft shares fell 0.6 % and Alphabet rose 1.3 % after the announcement.

The National Institute of Standards and Technology

The government seeks to identify AI-enabled threats—cyber-attacks, biosecurity risks, weapon-system misuse—before public release. The Pentagon’s parallel pact with seven AI firms aims to augment warfighter decision-making, reflecting a broader AI-integration strategy.

Al Jazeera (website)

CAISI Director Chris Fall said rigorous measurement science is essential for understanding frontier AI and its national-security implications, adding the collaborations help scale work. Microsoft’s Crampton noted CAISI adds technical expertise. A White House spokesperson said policy announcements will come directly from the President and that executive-order talks are speculative. The Defense Department said the agreements will augment warfighter decision-making.

The National Institute of Standards and Technology

Analysts note developers often provide models with reduced safety guardrails, raising misuse concerns. Anthropic’s refusal to lower guardrails for military use led the Pentagon to label it a supply-chain risk and sparked litigation. Draft legislation to codify CAISI’s authority has been introduced, but no bill has passed.

Google, Microsoft, and xAI agree to let US government test AI models before public release — OpenAI and Anthropic also on board after renegotiating deals with Washington

CAISI director Chris Fall is quoted as emphasizing “rigorous” measurement science, while a separate NIST release attributes a similar statement to “Chris Fell” and uses “vigorous.” No list identifies which models will be evaluated, and post-deployment review criteria remain undisclosed. Google and xAI declined comment, leaving testing protocols unknown.

Google DeepMind, Microsoft and xAI Sign Agreements for US National Security AI Testing

• “Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” — Chris Fall, CAISI Director

CNN

• “These expanded industry collaborations help us scale our work in the public interest at a critical moment.” — Chris Fall, CAISI Director

CNN

• “While Microsoft regularly tests its own models, CAISI offers additional “technical, scientific and national security expertise,” Microsoft Chief Responsible AI Officer Natasha Crampton said in a statement Google declined to comment further on the agreement.” — Natasha Crampton, Microsoft Chief Responsible AI Officer

CNN

• “Any policy announcement will come directly from the president. Discussion about potential executive orders is speculation.” — White House spokesperson

A White House AI-oversight working group will issue recommendations within weeks, and the administration may issue an executive order for a formal pre-release review. CAISI will continue post-deployment assessments and may broaden its mandate to cover additional frontier technologies. Ongoing Anthropic litigation and pending congressional bills indicate the AI-testing regulatory framework remains unsettled.