About a week ago, the Commerce Department’s Center for AI Standards and Innovation (CAISI) announced a deal with the AI companies Microsoft, xAI, and Google that allowed the government to inspect unreleased AI models before they’re released to the general public. Anthropic and OpenAI signed something similar back in 2024. However, the official announcement page has since disappeared from the CAISI website, raising concerns about transparency in government AI oversight.

The original announcement, dated May 5, 2026, stated: “Today, the Center for AI Standards and Innovation (CAISI) at the Department of Commerce’s National Institute of Standards and Technology announced new agreements with Google DeepMind, Microsoft and xAI. Through these expanded industry collaborations, CAISI will conduct pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security. These agreements build on previously announced partnerships, which have been renegotiated to reflect CAISI’s directives from the secretary of commerce and America’s AI Action Plan.”

That text is now only available via the Wayback Machine. The original URL now redirects to an error page and then to the main CAISI page on the Commerce Department website. As of Monday night, the redirect remained in place. The sudden removal has perplexed observers who follow federal AI policy, especially given the high profile of the companies involved — Google, Microsoft, and xAI (the company founded by Elon Musk).

The Significance of the Missing Page

The missing page is not just a technical glitch. It represents a potential gap in public accountability regarding how the federal government interacts with the most powerful AI labs. Pre-deployment evaluations are a cornerstone of the Biden-era AI executive order and subsequent policy frameworks. By having government scientists test frontier models before they reach the market, the aim is to catch dangerous capabilities — such as enhanced cyberattack tools or bioweapons planning — before they become widespread.

CAISI, established within the National Institute of Standards and Technology (NIST), is the agency tasked with creating voluntary standards and conducting evaluations. It was formalized in 2024 under the AI Safety and Security Board. The agreements with Google, Microsoft, and xAI were seen as a significant expansion of these efforts, which previously only included OpenAI and Anthropic. The inclusion of xAI, known for its Grok model, was particularly notable given Musk’s vocal criticism of AI safety doomsayers and his own stated goals of creating a maximally curious AI.

The announcement also came at a time when the United States is racing to set global AI standards in the face of competition from China and the European Union's AI Act. Any perceived lack of transparency could undermine U.S. credibility in international negotiations.

Background on Government AI Oversight

Federal oversight of AI has grown in fits and starts. The first voluntary commitments from AI companies came in July 2023, when the Biden administration announced agreements with seven leading AI firms, including Google, Microsoft, and OpenAI. Those commitments included red-teaming and sharing safety information with the government. However, the creation of CAISI and the formalization of pre-deployment evaluations represented a stronger, institutionalized approach.

Under the Trump administration (which had returned to power in January 2025), AI policy took a turn toward deregulation, but the Commerce Department still maintained CAISI as a technical agency. The May 5, 2026 announcement was the first major public confirmation of ongoing evaluations under the new administration. The language in the announcement referenced “America’s AI Action Plan,” a term from the current administration’s policy document released in early 2026, which called for public-private partnerships to maintain U.S. leadership in AI without heavy-handed regulation.

The disappearance of the page could signal a policy shift, a bureaucratic error, or even an intentional removal due to leaked details that the government did not want public. Without official comment, the motive remains unclear.

What the Archived Announcement Said

The archived version of the page, saved by the Wayback Machine, contains more than just the opening paragraph. It continues: “These agreements support information-sharing and ensuring a clear understanding in government of AI capabilities and the state of international AI competition.” It also noted that the evaluations would be conducted by NIST researchers using the government’s own test beds and that results would be shared with the companies to improve safety, but not necessarily made public.

The lack of public reporting on specific findings has been a point of criticism. Advocates for transparency argue that without open publication, there is no way for independent researchers or the public to verify that the evaluations are rigorous. The page’s removal only amplifies those concerns.

Possible Reasons for Removal

There are several plausible explanations for why the page vanished. One is technical: a website migration or URL change could have accidentally broken the link. However, the fact that the page is gone and replaced by a redirect suggests an intentional decision by the site administrators.

Another possibility is that the announcement contained sensitive dates or details about upcoming evaluations that the administration now considers inappropriate for public consumption. For instance, it might have included timelines for testing that, if known, could tip off adversaries or trigger market speculation. Alternatively, the agreements themselves may have been renegotiated or withdrawn, and the government did not want the outdated page to remain.

A third, less charitable explanation is that the administration wants to downplay the extent of its AI oversight in order to signal a friendlier posture toward tech companies. This would align with the current administration’s general skepticism of federal regulation. However, such a move would likely face backlash from AI safety advocates who argue that self-regulation by companies is insufficient.

The timing is also curious. The disappearance comes just days before a scheduled congressional hearing on AI safety where Commerce Department officials were expected to testify. It is possible that the page was taken down to avoid providing ammunition to lawmakers demanding stricter oversight.

Reactions and Next Steps

Journalists and transparency organizations have begun questioning the removal. The nonprofit Project on Government Oversight (POGO) sent an inquiry to the Commerce Department, but as of Monday evening had received no reply. Freedom of Information Act requests are likely to follow.

For now, the only public evidence of the agreements exists in archived snapshots and in the memory of those who read the page before it vanished. Researchers who track AI policy are urging the government to restore the information or issue a new statement explaining the change.

The companies involved — Google, Microsoft, and xAI — have not commented on the missing page. They may be waiting for the government to take the lead in communicating about the agreements. Meanwhile, the AI community watches to see whether this is a minor administrative glitch or a sign of a broader retreat from transparency.

The episode highlights the fragile nature of digital government communications. When web pages are removed without explanation, it erodes trust and fuels speculation. For an issue as consequential as AI safety, clarity is essential.

Source: Gizmodo News