Commerce Department Removes AI Security Test Announcement From Website

The Commerce Department removed details from its website about an agreement with Google, xAI and Microsoft to let federal scientists test new artificial intelligence models for security flaws, changing the public record of a plan meant to catch risks before those systems reached users.

The department had announced on May 5 that the three companies would give the government access to models before public release so officials could look for vulnerabilities. The stated aim was broad and consequential: identify weaknesses that could be exploited in cyberattacks or in military misuse before frontier systems were widely deployed.

By Monday afternoon in Washington, the link that had pointed to the announcement no longer worked. It first produced a message saying the page could not be found, then later redirected to the Center for AI Standards and Innovation website. The shifting status of the page mattered because the underlying arrangement involved some of the most powerful names in AI and a direct role for the federal government in testing models before they reached the public.

The removal came as concern has intensified inside the U.S. government over the national security risks posed by advanced AI systems. That made the disappearance of the announcement more than a routine website cleanup. It raised a basic accountability question: whether the information was simply moved under a specialized AI standards office, or whether a public-facing statement about the testing deal was deliberately scrubbed from view.

It was not immediately clear why the page was taken down, and spokespeople for the Commerce Department and the White House did not immediately respond. That silence left open the question of who ordered the change and what it signals about how closely the government wants the public to see its coordination with major AI companies.

The episode lands at a moment when federal officials are trying to balance rapid AI development with tighter oversight of systems that can be put to use quickly and at scale. The agreement with Google, xAI and Microsoft had promised a rare window into model testing before release. The disappearance of the announcement now puts the government’s transparency under scrutiny just as oversight of frontier AI grows more important, not less.