The Commerce Department has removed from its website the details of an agreement under which Google, Microsoft, and Elon Musk’s xAI had agreed to submit new AI models to government scientists for security testing before public release, according to Reuters.
Published on May 5, the now-removed page described an arrangement in which Google, Microsoft, and xAI would give government scientists early access to frontier models so they could be assessed for vulnerabilities ranging from cyberattack potential to military misuse before those systems reached the public. Anyone who visited the link on Monday found it had gone dark, displaying an error that read “Sorry, we cannot find that page.” The URL was subsequently rerouted to the homepage of the Center for AI Standards and Innovation, the federal body that oversees the testing program.
No explanation for the removal was offered publicly, and Reuters noted that neither Commerce Department spokespeople nor the Trump White House returned its requests for comment.
On May 5, CAISI unveiled agreements with all three companies — housed, like the rest of CAISI, inside the National Institute of Standards and Technology within the Commerce Department — under which federal evaluators would be granted access to advanced AI models ahead of their public launch. CNBC reported that the new deals extended a framework CAISI had already established with OpenAI and Anthropic the previous year. CAISI was formerly known as the U.S. AI Safety Institute before being renamed and refocused under the Trump administration.
Microsoft said its agreement with CAISI would include collaboration on methodologies for adversarial assessments — testing AI systems for unexpected behaviors, misuse pathways, and failure modes. The company said it would also work with the U.K.’s AI Security Institute on research related to frontier safety and security.
Neither Microsoft, Google, nor xAI responded to Reuters’ requests for comment on the deletion. According to The Next Web, the fact that the broken link now routes to CAISI’s site rather than simply going dark hints that the underlying testing arrangements have not necessarily been terminated.
The deletion comes as the Trump administration has been weighing a broader AI oversight structure. The White House has been considering an executive order that would create a working group of tech executives and government officials to study pre-deployment review processes for AI models. The discussions marked a shift for an administration that moved quickly to eliminate Biden-era AI safety requirements after President Donald Trump took office.
The Next Web also noted that a number of officials within the federal government have challenged the logic of the program itself, warning that early government access to powerful, unreleased AI systems could draw the attention of foreign intelligence services looking to exploit that access.
A person familiar with the deliberations, who requested anonymity given the sensitivity of the matter, told CNBC the group was expected to draw in officials from across the government alongside executives from major technology companies. On the question of the proposed working group, CNBC’s source indicated that deliberations inside the White House had not yet concluded, with an executive order remaining one possible vehicle for formalizing it. Asked for comment, the White House directed CNBC to await any announcement from President Donald Trump himself.