The US Commerce Department deletes website details of Microsoft, Google, and xAI security-test deal

The US Commerce Department has removed from its website the details of an agreement under which Microsoft, Google, and xAI agreed to submit new AI models to government scientists for security testing before release, Reuters reported on Monday.

The original page, posted on 5 May, said the three companies would hand over their frontier AI systems to the department’s testing team to be reviewed for cyberattack vulnerabilities, military-misuse risk and national-security flaws before public deployment.

By Monday afternoon Washington time, the link returned a “Sorry, we cannot find that page” error message; it was subsequently redirected to the website of the Center for AI Standards and Innovation, the government body that runs the tests.

The Center, the successor body to the US AI Safety Institute (AISI), is housed within the National Institute of Standards and Technology (NIST), part of the Department of Commerce.

The renaming and refocusing followed an executive order that scaled back the previous administration’s AI-safety architecture and reframed the institute’s mission around standards and industry coordination rather than safety evaluation.

Reuters reported that neither the Commerce Department nor the Trump White House responded immediately to requests for comment on why the page was deleted. There is no public statement from Microsoft, Google, or xAI on the change.

The May 5 announcement had been read at the time as a notable commitment by the three companies to pre-deployment government review, and as a sign of growing federal concern about national-security risks posed by powerful AI systems.

The deal followed the Trump administration’s removal of Anthropic from a Pentagon AI contract over alleged safety-related constraints; Anthropic was not named as a participant in the Commerce Department testing programme.

The deletion does not necessarily mean the programme has been cancelled. The Center for AI Standards and Innovation continues to operate, and the redirected webpage suggests the relationship between the agencies and the three companies remains in place at an operational level.

Several federal officials have, however, publicly questioned the wisdom of giving the government access to frontier AI models pre-release, because such access could become a target for nation-state cyber-espionage.

The story matters most as a signal. The Commerce Department’s willingness to remove a positive AI-safety announcement from its public-facing website, without explanation, will be read by both critics and supporters of US AI policy as evidence of internal disagreement about how the government should engage with frontier AI labs.

Industry observers had treated the original 5 May announcement as a stable element of the new administration’s AI-policy posture.

Microsoft, Google and xAI did not respond to Reuters’ requests for comment. Anthropic, OpenAI, Meta and other large model providers were not part of the original announcement and have not commented on the deletion.

The Center for AI Standards and Innovation’s website, where the redirect now points, contains general information on its programme but does not currently include the specifics of the pre-release testing arrangement that were on the deleted Commerce page.