The U.S. Commerce Department has taken down online details about its collaborative testing arrangement with Google, xAI and Microsoft that would allow government scientists to examine new artificial intelligence models for security flaws before those models are released publicly.
The web address that formerly pointed to the Commerce Department announcement now returns an error. Reporters in Washington observed the page displaying the message, "Sorry, we cannot find that page." Subsequently, attempts to access the removed content redirected users to the Center for AI Standards and Innovation's site. That government office is identified as the entity responsible for carrying out the model testing.
In a public notice earlier this week, the Commerce Department said participating companies would provide their newer AI models to the government prior to public deployment. The stated purpose of that advance access is to give government scientists the opportunity to test the models for security vulnerabilities.
Officials have expressed growing concern across the federal government about the national security implications of increasingly powerful AI systems. The announcement framed early access as a means to detect a spectrum of potential threats, specifically noting risks that include cyberattacks and the potential for military misuse. The department’s statement cited those broad categories of threats as the focus of the planned testing.
The announcement singled out an example of concern by name. The report referenced Anthropic's Mythos as one of the powerful AI systems that have drawn attention from policymakers and security officials. The Commerce Department did not, however, provide an explanation at the time for why the webpage with the testing details was later removed from its site.
The Center for AI Standards and Innovation was identified in the online redirection as the government organization charged with administering the tests, but the sequence of the announcement followed by the page removal leaves the public record on the timing and rationale for the change incomplete.
Context note: The department's earlier announcement described the process under which companies would hand over new models before public deployment so that government scientists could screen them for security weaknesses. Beyond the fact of the announcement and the subsequent removal of the webpage, officials have not provided a public reason for the deletion.