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 for these agreements has since disappeared from the CAISI website, leaving journalists, researchers, and the public scrambling to understand what happened and why.
The original URL now redirects to the main CAISI page on the Commerce Department website, displaying an error message that reads, "Sorry, we cannot find that page." As of this writing, the redirect remains active. The disappearance was first noticed by Reuters on Monday afternoon, and it has since sparked discussions about transparency in government AI oversight and the stability of public-facing commitments from federal agencies.
What the Announcement Said
The archived version of the page, preserved by the Wayback Machine, provides a detailed description of the agreements. It states: "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."
The archived page also noted that "these agreements support information-sharing" and help in "ensuring a clear understanding in government of AI capabilities and the state of international AI competition." The phrasing suggests that the government views such pre-deployment evaluations as essential not only for safety but also for maintaining a competitive edge against other nations, particularly China and the European Union, which have their own aggressive AI strategies.
The Significance of Pre-Deployment Evaluation
Pre-deployment evaluation of AI models is a relatively new but increasingly critical aspect of AI governance. Unlike traditional software, frontier AI models—such as large language models (LLMs) and multimodal systems—can exhibit emergent behaviors that are not fully predictable even by their creators. By allowing government researchers to inspect models before public release, these agreements aim to catch potential risks related to bias, misinformation, cybersecurity vulnerabilities, and even existential threats before they reach millions of users.
Google DeepMind, the AI research lab behind Gemini, has been at the forefront of developing advanced models that integrate reasoning, code generation, and multimodal inputs. Microsoft, through its partnership with OpenAI and its own development of Copilot and Azure AI services, has deployed AI across enterprise and consumer products. xAI, founded by Elon Musk, has released the Grok model, which is designed to be more direct and less filtered than other chatbots. Each of these models carries unique risks and benefits, making government oversight both necessary and challenging.
The agreements with Anthropic and OpenAI in 2024 set a precedent. Those earlier deals were lauded by AI safety advocates as a step toward responsible innovation. However, they also raised concerns about industry capture, lack of enforcement mechanisms, and the potential for confidential information to be mishandled. The new agreements with Google, Microsoft, and xAI were supposed to build on that foundation, incorporating lessons learned from the first wave of evaluations.
Why Did the Page Disappear?
The exact reason for the page's removal remains unclear. The White House and the Department of Commerce did not immediately respond to requests for comment. Several theories have emerged among policy analysts.
One possibility is that the page was taken down due to a technical error or website reorganization. The CAISI website has undergone several updates since its launch, and it is possible that the announcement was inadvertently caught in a content migration. However, the redirect to the main page suggests a deliberate action rather than a simple broken link.
Another theory is that the agreements themselves are being renegotiated or that new details have emerged that require the announcement to be revised. The original text mentioned that the partnerships had been "renegotiated to reflect CAISI's directives." It is plausible that the government wants to update the language to reflect the latest terms, and the removal is a temporary measure while a new version is drafted.
A more concerning possibility is that the page was removed under pressure from the companies involved or from political actors who are skeptical of government regulation of AI. The AI industry has a mixed record on transparency. While companies like Google and Microsoft publicly support responsible AI development, they have also lobbied against strict regulatory frameworks. If the page removal was intended to limit public scrutiny, it would represent a significant setback for transparency.
Historical Context and the AI Action Plan
The AI Action Plan referenced in the announcement is a White House initiative that outlines the administration's strategy for managing AI risks while promoting innovation. It includes calls for voluntary commitments from AI companies, establishment of safety institutes, and investment in AI research. The plan has been criticized by some for being too reliant on industry self-regulation and for lacking binding enforcement.
CAISI itself was created within NIST to serve as a hub for AI standards and evaluation. Its mission includes developing technical guidelines for trustworthy AI, conducting evaluations of frontier models, and collaborating with international partners. The center has been involved in projects like the AI Risk Management Framework and the evaluation of AI systems used in critical infrastructure. The agreements with Google, xAI, and Microsoft were among its highest-profile initiatives.
Comparatively, the United Kingdom has established its own AI Safety Institute, which has signed agreements with multiple AI companies to allow pre-release access to models. The EU's AI Act imposes mandatory requirements on high-risk AI systems. The missing page in the U.S. could be seen as a sign of inconsistency in the American approach, especially at a time when other nations are moving decisively to codify AI rules.
Public Reaction and Media Coverage
News of the missing page quickly spread across social media and tech news outlets. Gizmodo was one of the first to cover the story, noting that the original URL led to an error page. The article highlighted the irony that a page promoting transparency and information-sharing had itself become inaccessible.
Comments from readers on the original article reflected a mix of frustration and skepticism. Some accused the government of hiding information, while others speculated that the removal was bureaucratic incompetence rather than malice. The incident also drew attention to the broader issue of how government agencies handle digital records and commitments. In an age of rapid policy changes and website reboots, ensuring that official announcements remain accessible is a fundamental component of good governance.
Journalists have used the Wayback Machine to preserve the original content, but the inability to access the page on the government's own site undermines public trust. If the agreements are still in effect, citizens have a right to know the details. If they have been suspended or altered, that, too, should be publicly announced.
The Role of the Companies Involved
Google DeepMind, Microsoft, and xAI have not publicly commented on the disappearance of the page. Each company has its own track record regarding government cooperation. Google has long engaged with regulators on AI ethics and has published transparency reports. Microsoft has committed to responsible AI principles and has established a dedicated AI governance office. xAI, being a younger and more controversial player, has been more guarded about its model evaluations.
Elon Musk, the founder of xAI, has been an outspoken critic of AI regulation while also warning about AI's existential risks. His company's participation in the CAISI agreements was seen as a positive step toward bipartisan collaboration on AI safety. If the page removal leads to confusion or delays in the evaluation process, it could damage the credibility of the entire program.
Furthermore, the renegotiation mentioned in the original announcement suggests that the terms of the agreements have evolved. It is possible that the companies sought changes to the confidentiality clauses, the scope of evaluations, or the timeline for public release. Without access to the original or revised text, outsiders can only speculate.
Implications for AI Safety and Oversight
The disappearance of the page has broader implications for the field of AI safety. Proponents of strong government oversight view pre-deployment evaluations as a vital tool for preventing harm. Without a transparent and publicly accessible record of such agreements, it becomes harder to hold both the government and the companies accountable.
Some experts argue that the evaluations themselves may be less effective if the details of the process are not shared with the public. The AI community relies on open research and peer review to identify flaws and improve safety measures. If the government's evaluations are conducted in secret, the lessons learned may not benefit the wider field. The missing page raises questions about whether CAISI intends to operate in a glass-box or black-box manner.
On the other hand, there are legitimate reasons for confidentiality. Pre-deployment evaluations often involve proprietary information about model architectures, training data, and internal safety tests. Revealing too much could aid malicious actors in finding vulnerabilities or enable competitors to copy innovations. Striking the right balance between transparency and security is a perennial challenge.
The incident also highlights the fragility of government websites as repositories of critical policy documents. Unlike static PDFs or press releases, webpages can be changed or removed without notice. For scholars, journalists, and the public, this creates a constant need for archiving. The Wayback Machine and similar tools have become essential for preserving digital history, but they are not infallible. The original page might have been removed before a snapshot was taken, or the snapshot might miss interactive elements such as embedded documents.
Looking Ahead
As of now, the CAISI website continues to redirect the original URL to the main page. No new announcement has been posted to replace the missing one. The White House and Commerce Department have not provided a timeline for when the information might be restored or updated.
For those following AI policy, this episode serves as a reminder that digital governance requires more than just crafting policy—it requires maintaining a consistent and accessible public record. The missing page may be a simple oversight, or it may signal deeper tensions within the administration about how to handle the rapidly evolving AI landscape. Either way, it underscores the need for robust transparency mechanisms that survive administrative changes and technical glitches.
The agreements with Google, xAI, and Microsoft represent a major step forward in the government's ability to assess frontier AI models. But if the public cannot even find the announcement, the benefits of that step are diminished. Restoring the page—or issuing a clear explanation for its removal—should be a priority for the Commerce Department. Until then, the mystery of the missing page will continue to cast a shadow over the government's AI vetting efforts.
Source: Gizmodo News