About a week ago, the Commerce Department’s Center for AI Standards and Innovation (CAISI) announced a significant step in AI governance: agreements with Google DeepMind, Microsoft, and xAI that would allow the government to inspect unreleased AI models before they reach the general public. The announcement, dated May 5, 2026, detailed how these expanded industry collaborations would enable CAISI to conduct pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security. It also noted that these agreements built on previously announced partnerships, which had been renegotiated to reflect CAISI’s directives from the secretary of commerce and America’s AI Action Plan.
However, that announcement page has now vanished from the CAISI website. Reuters first reported the disappearance on Monday afternoon, noting that the original URL initially returned an error page saying “Sorry, we cannot find that page,” and later redirected to the main CAISI page on the Commerce Department website. As of Monday night, the URL still redirects to the CAISI page, effectively removing the original announcement from public view. The only remaining copy of the text is preserved through the Wayback Machine, which captured the full announcement before it was pulled.
The missing page raises immediate questions about transparency in the government’s approach to artificial intelligence oversight. The agreements with Google, Microsoft, and xAI represent a key component of the Biden administration’s broader AI strategy, which emphasizes voluntary commitments and pre-deployment testing to mitigate risks from advanced AI systems. Similar agreements with Anthropic and OpenAI were signed back in 2024, setting a precedent for voluntary government access to frontier models. The disappearance of the announcement page could be a technical glitch, but given the sensitive nature of the information—and the fact that it involves direct government access to proprietary AI models—the lack of a clear explanation is troubling.
To understand the significance of these agreements, it helps to look at the broader landscape of AI safety efforts. Since the release of ChatGPT in late 2022, governments worldwide have scrambled to establish frameworks for evaluating and regulating advanced AI systems. In the United States, the NIST (National Institute of Standards and Technology) has been at the forefront, developing the AI Risk Management Framework and later establishing CAISI to focus specifically on frontier AI—models that are at the cutting edge of capability and may pose novel risks. The pre-deployment evaluations announced in May 2026 would allow CAISI researchers to test models for dangerous capabilities, such as cybersecurity vulnerabilities, persuasion manipulation, or autonomous replication, before they are deployed to millions of users.
The agreements with Google DeepMind, Microsoft, and xAI mark an expansion of the government’s access beyond the original participants (Anthropic and OpenAI). xAI, founded by Elon Musk, has positioned itself as a competitor to companies like OpenAI and Google, and its inclusion in these agreements signals that the government is seeking broad coverage across the AI industry. Microsoft, as a major investor in OpenAI and developer of its own AI models, also plays a dual role. The archived announcement stated that “these agreements support information-sharing” and “ensuring a clear understanding in government of AI capabilities and the state of international AI competition.”
Yet the very page that contained those words is now missing. The timing is curious: the announcement went live on May 5, and by May 11—when Reuters and later Gizmodo began investigating—the page was gone. Gizmodo requested comment from the White House and Commerce Department on Monday evening but did not immediately receive a reply. Without an official statement, speculation ranges from a routine website refresh to a deliberate removal due to some undisclosed concern about the content of the agreement or the wording of the announcement.
This incident is not an isolated one. Government websites often undergo changes, but the disappearance of a major policy announcement is rare. In the realm of AI, where trust and transparency are already fragile, such actions can undermine public confidence. Critics have long argued that voluntary agreements lack enforcement mechanisms and that the government should be more transparent about how it evaluates and interacts with AI companies. The vanishing announcement may fuel calls for statutory regulation rather than voluntary pacts.
Moreover, the content of the announcement itself suggests that these agreements are part of a larger, ongoing process. The renegotiation mentioned in the text implies that earlier partnerships were updated to align with new directives. What those directives are, and how they change the terms of access, remains unclear. The removal of the page obscures not only the current state of these agreements but also the evolution of the government’s AI strategy.
The broader context includes international dimensions. The United States is competing with China and the European Union in setting standards for AI governance. The EU’s AI Act, passed in 2024, imposes strict requirements on high-risk AI systems, while China has its own regulatory framework emphasizing state control. The US approach has favored industry self-regulation and voluntary commitments. The loss of the announcement page could be seen as a setback for transparency, even if unintentional, and may give ammunition to those arguing for more robust oversight.
In the days ahead, expect further scrutiny from journalists, watchdog groups, and lawmakers. The Wayback Machine currently preserves the original announcement, but if the government does not restore the page or provide an explanation, the incident may become a flashpoint in the ongoing debate about AI accountability. For now, the official record is incomplete, and the public is left wondering why an announcement that seemed to represent a positive step in AI safety has been pulled from the public eye.
The disappearance also highlights the fragility of digital information. Government websites are supposed to be stable sources of policy documents, but technical errors, security concerns, or administrative decisions can remove information overnight. This reinforces the importance of independent archiving efforts, such as the Wayback Machine, as a check on institutional memory loss. As AI continues to evolve, the public’s ability to track government actions will depend on both official transparency and unofficial preservation.
Ultimately, the missing page is a small but telling episode in the larger story of AI governance. It demonstrates that even routine announcements can become controversial when they vanish without explanation. The government’s willingness to share information about its AI dealings with major tech companies is crucial for maintaining trust. Until the page is restored or an official explanation is provided, the incident will remain a notable footnote in the history of US AI policy.
Source: Gizmodo News