The US Commerce Department removes security test information for Microsoft, Google, and xAI.

      The US Commerce Department has taken down from its website information regarding an agreement where Microsoft, Google, and xAI had consented to present new AI models to government scientists for security evaluation prior to their release, as reported by Reuters on Monday.

      The page, which was originally published on May 5, indicated that the three companies would submit their advanced AI systems to the department's testing team for an assessment focused on vulnerabilities to cyberattacks, potential military misuse, and national security issues before public launch.

      By Monday afternoon, Washington time, the link displayed a “Sorry, we cannot find that page” error and was then redirected to the site of the Center for AI Standards and Innovation, the governmental organization in charge of conducting the tests.

      The Center, which replaced the US AI Safety Institute (AISI), operates within the National Institute of Standards and Technology (NIST), a component of the Department of Commerce. The rebranding and shifting of focus occurred following an executive order that diminished the previous administration's AI-safety framework, redirecting the institute's mission towards standards and industry collaboration rather than safety evaluations.

      Reuters noted that neither the Commerce Department nor the Trump White House immediately responded to inquiries about the rationale behind the page's removal. There has been no public comment from Microsoft, Google, or xAI regarding this change.

      The announcement on May 5 was interpreted at the time as a significant commitment from the three companies to allow for government review prior to deployment, reflecting increasing federal worries about national security risks associated with advanced AI technologies.

      This agreement followed the Trump administration's decision to remove Anthropic from a Pentagon AI contract over concerns related to safety; however, Anthropic was not mentioned as a participant in the Commerce Department testing initiative.

      The deletion does not necessarily imply the cancellation of the program. The Center for AI Standards and Innovation continues its operations, and the redirected webpage indicates that the relationship between the agency and the three companies remains intact at an operational level.

      Nonetheless, several federal officials have publicly raised concerns about the wisdom of granting the government access to advanced AI models before their release, as such access could become a target for cyber-espionage by nation-states.

      This incident is significant primarily as a signal. The Commerce Department's decision to remove a positive announcement regarding AI safety from its public website, without any explanation, will likely be interpreted by both critics and advocates of US AI policy as indicating internal disagreements on how the government should interact with leading AI laboratories.

      Observers in the industry had regarded the initial announcement from May 5 as a reliable component of the new administration’s AI policy stance. Microsoft, Google, and xAI did not respond to Reuters’ requests for comments. Other major model providers such as Anthropic, OpenAI, and Meta were not included in the original announcement and have not commented on the page's removal.

      The website of the Center for AI Standards and Innovation, to which the redirect now leads, provides general information about its program but currently lacks the details of the pre-release testing arrangement that were available on the now-deleted Commerce page.