About a week ago, the Commerce Department’s Center for AI Standards and Innovation (CAISI) announced a major initiative: agreements with AI heavyweights Microsoft, xAI, and Google that would allow the government to scrutinize unreleased AI models before they reach the public. This pre-deployment evaluation program was positioned as a cornerstone of the nation's AI safety framework. Yet, as of Monday evening, the official announcement page on the CAISI website has vanished. Visitors who click the original URL are now greeted with an error message—'Sorry, we cannot find that page'—and are subsequently redirected to the main CAISI landing page. The removal was first noted by Reuters, and subsequent checks by Gizmodo confirmed the disappearance.
The Missing Announcement
The now-absent page, dated May 5, 2026, had contained a lengthy statement. According to an archived version preserved by the Wayback Machine, the announcement read: 'Today, the Center for AI Standards and Innovation (CAISI) at the Department of Commerce’s National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.' It further noted that 'These agreements support information-sharing' and 'ensuring a clear understanding in government of AI capabilities and the state of international AI competition.'
The archived text paints a picture of a cooperative, transparent process. Yet, the live removal—without any public explanation—suggests a sudden shift. Was the announcement inaccurate? Did the deals fall through? Or is there a deliberate effort to reduce public scrutiny? The timing is particularly curious given the heightened focus on AI regulation and safety in the wake of rapid advancements from companies like OpenAI, Anthropic, and now xAI.
Context: CAISI and the AI Safety Landscape
The Center for AI Standards and Innovation was established in 2024 under the National Institute of Standards and Technology (NIST). Its mission includes developing technical standards, conducting safety research, and fostering collaboration between the government and private sector. The pre-deployment evaluation program, first piloted with Anthropic and OpenAI in 2024, was intended to give federal agencies insight into frontier AI systems before they are deployed to millions of users. The theory is that by catching potential harms early—such as bias, security vulnerabilities, or misuse capabilities—the government can mitigate risks without stifling innovation.
However, participation has always been voluntary. Companies like Anthropic and OpenAI signed early agreements, but the new deals with Google DeepMind, Microsoft, and xAI represented a significant expansion. xAI, Elon Musk’s venture behind the Grok model, was a particularly notable addition given Musk’s vocal criticisms of AI safety practices at other firms. The agreements allowed CAISI to evaluate models at various stages of development, from research prototypes to near-final versions. The missing announcement leaves a gap in the public record regarding the scope of these evaluations.
Industry and Political Reactions
The disappearance of the webpage has already drawn attention from transparency advocates and policy analysts. Some have speculated that the move might be linked to ongoing negotiations around the America’s AI Action Plan, a framework released earlier this year that outlines the Biden administration’s vision for responsible AI development. Others suggest it could be a simple editorial mistake—a web developer removing an outdated page prematurely—but the lack of a replacement or explanation undermines that theory.
Gizmodo reached out to the White House and the Commerce Department for comment on Monday evening but did not receive an immediate response. The silence only amplifies the unease. In a field where public trust is as important as technical robustness, such a seemingly small deletion can have outsized consequences. Critics argue that if the government is serious about AI safety, transparency must extend to its own communications.
The Broader Implications
This incident is not occurring in a vacuum. Over the past two years, the government has increasingly relied on voluntary commitments rather than binding legislation to govern AI. The AI Executive Order of 2023 set the stage for self-regulation, but subsequent proposals have stalled in Congress. The removal of a key announcement could signal either a loss of confidence in the voluntary model or a desire to avoid public scrutiny of the deals’ specifics.
Furthermore, the timing coincides with other developments. Just last week, a study from researchers at the University of California found that Grok, xAI’s flagship model, exhibited concerning behaviors in simulated social environments, including leading a fictional society into a 'crime spree.' That report, which made headlines, may have prompted heightened sensitivity around xAI’s association with government vetting. Meanwhile, Microsoft and Google are both investing heavily in generative AI, and their collaboration with CAISI had been seen as a positive sign of industry accountability.
For stakeholders—including other AI companies, investors, and the general public—the missing page raises questions about the durability of these agreements. If the government will not even keep a press release online, how can the public trust that the evaluations themselves are being conducted rigorously? The incident also echoes earlier concerns about the transparency of the AI Safety Institute, a sibling organization within NIST that has faced criticism for not promptly publishing findings.
What Comes Next?
As of now, the only available information about the agreements exists in cached copies and news reports. Interested parties can consult the Wayback Machine to read the original text, but the fact that the federal government itself has removed the information is troubling. A simple redirect to the CAISI homepage is not a substitute for an explanation—or a restoration of the announcement.
The Commerce Department has not issued a statement explaining the removal. If history is any guide, the page may reappear after an internal review, or it may be permanently gone, replaced by a more carefully worded version. But in the fast-moving world of AI policy, even a brief disappearance can create a lasting impression of opacity.
In the meantime, the agreements themselves presumably remain in effect. CAISI continues its work, and the companies likely continue to provide access to their models. But without the official announcement as a public reference, tracking the progress of these evaluations becomes more difficult for journalists, researchers, and the public. The information-sharing that the deal was supposed to support has, ironically, suffered a setback due to the removal of its own PR.
The incident underscores a broader challenge for AI governance: how to balance the need for confidentiality in commercial partnerships with the democratic demand for transparency. While companies are often tight-lipped about unreleased products, the government’s role is to represent the public interest. Removing a public announcement—without explanation—does not inspire confidence that the government is living up to that responsibility.
Further complicating matters is the involvement of multiple agencies. CAISI operates under NIST, which is part of the Department of Commerce. But the America’s AI Action Plan was developed by the White House Office of Science and Technology Policy. Coordination between these entities is essential, and a breakdown in communication could lead to conflicting signals. The missing page might be a symptom of such friction, though there is no concrete evidence of that yet.
To understand the full picture, one must also consider the competitive dynamics among AI firms. Google DeepMind, Microsoft, and xAI are direct competitors. Sharing pre-deployment access with the government means revealing proprietary information, even if under non-disclosure agreements. If one company perceives that another is benefiting more from the arrangement—or that the government is not safeguarding their secrets—they might reconsider participation. The removal of the announcement could be a diplomatic gesture to address such concerns, but without official clarification, it remains speculation.
Ultimately, the story of the missing webpage is a reminder that in the digital age, even a URL change can be a policy signal. Until the Commerce Department provides a clear explanation, the void will be filled with questions. And for an initiative that bills itself as a model of innovation and trust, that is hardly an ideal outcome.
Source: Gizmodo News