In a development that has sparked concerns about government transparency, a webpage announcing new AI safety agreements between the Commerce Department and three of the world’s largest AI companies—Google DeepMind, xAI, and Microsoft—has gone missing from the official website of the Center for AI Standards and Innovation (CAISI). As of late Monday, the original URL for the May 5, 2026 announcement now redirects to the main CAISI page on the Department of Commerce website, leaving no trace of the original content besides archived versions preserved by the Wayback Machine.
The missing page had detailed a significant expansion of the government's role in evaluating advanced AI models before their public release. Under these agreements, CAISI would conduct pre-deployment evaluations and targeted research to assess frontier AI capabilities and advance the state of AI security. The announcement emphasized that these collaborations were built on previously negotiated partnerships, which had been renegotiated to align with directives from the secretary of commerce and the America’s AI Action Plan.
What the Announcement Contained
An archived version of the page, captured by the Internet Archive’s Wayback Machine on May 5, 2026, reveals the full text of the announcement. It 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.”
The archived page also included a statement that “these agreements support information-sharing, ensuring a clear understanding in government of AI capabilities and the state of international AI competition.” This language underscores the dual purpose of the partnerships: both to enhance safety and to maintain U.S. competitiveness in the global AI race.
The Disappearance
The missing page was first noticed by Reuters, which reported on Monday afternoon that the original URL returned a “Sorry, we cannot find that page” error and later redirected to the main CAISI page. By Monday evening, the redirect was still in place. Gizmodo attempted to reach the White House and the Commerce Department for comment but did not receive an immediate response. The timing of the page’s removal is unclear, but it appears to have happened sometime between May 5 and May 11, when the article reporting on it was published.
This is not the first time that government AI safety announcements have been subject to scrutiny. In 2024, similar agreements were announced with OpenAI and Anthropic, and those pages remain accessible. The disappearance of the newer agreements raises questions about whether the content was deliberately removed or if it was caught in a routine website reorganization. However, the lack of a redirect to an updated version or a notice explaining the change fuels speculation about the government’s commitment to transparency in AI policymaking.
Historical Context of AI Safety Agreements
The concept of pre-deployment evaluations for AI models emerged as a key recommendation from the Biden administration’s 2023 Executive Order on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. That order called for leading AI developers to share safety test results with the U.S. government. Following the order, the Department of Commerce established CAISI within NIST to coordinate these efforts. The initial agreements with OpenAI and Anthropic in late 2024 set a precedent for voluntary commitments that allowed government researchers to test models for risks such as misuse, bias, and national security threats.
The new agreements with Google DeepMind, Microsoft, and xAI represented a significant expansion. These companies are among the most prominent players in frontier AI development. Google DeepMind has pioneered advances in reinforcement learning and generative AI, Microsoft has integrated OpenAI’s models into its Azure cloud platform, and xAI has pushed the boundaries of large language models with its Grok system. The inclusion of xAI, led by Elon Musk, was particularly notable given Musk’s public criticism of AI safety measures and his own advocacy for regulation.
The agreements were also symbolically important because they demonstrated bipartisan continuity in AI policy. The America’s AI Action Plan, referenced in the announcement, was a legislative initiative aimed at maintaining U.S. leadership in AI while mitigating risks. The plan had received support from both Democrats and Republicans, highlighting the rare area of consensus in an otherwise divided political landscape.
Implications of the Page’s Removal
The sudden disappearance of the announcement could be interpreted in several ways. One possibility is that the page was taken down inadvertently during a site update and will be restored. Another is that the content needed revision due to factual errors or changed circumstances. However, the fact that no corrected version has appeared suggests a more deliberate decision.
Critics argue that the removal hampers public accountability. Without a stable record of the agreements, it becomes harder for civil society groups, researchers, and journalists to track the government's activities in AI oversight. The reliance on the Wayback Machine for verification underscores the fragility of digital government records. This incident adds to a growing list of examples where government websites have altered or removed content without explanation, including pages on climate change, public health, and civil rights.
Supporters of the removal might point to the sensitive nature of pre-deployment evaluations. Detailed information about the testing protocols or the specific models involved could potentially be exploited by malicious actors. But the announcement itself was high-level and contained no technical specifics that would compromise security. Moreover, the redirect does not lead to a “page not found” but to the main CAISI site, which still contains other information about the center’s mission but lacks any mention of the agreements.
The Broader AI Landscape
The missing page comes at a time when AI safety is increasingly in the forefront of public debate. Just last month, researchers from leading labs published a paper warning about the risks of advanced AI systems, including potential loss of human control. The U.S. government has been working to establish frameworks for testing and monitoring AI, but progress has been slow compared to the rapid pace of commercial development.
Other countries, notably the European Union and the United Kingdom, have moved faster to implement binding regulations. The EU’s AI Act, which entered into force earlier this year, imposes strict requirements on high-risk AI systems, including mandatory conformity assessments. The UK has established its own AI Safety Institute, which conducts evaluations similar to those envisioned by CAISI. The disappearance of the U.S. announcement could be seen as a setback in the international competition for AI governance leadership.
Meanwhile, the companies themselves have continued to develop and deploy new models. Google DeepMind recently unveiled a new version of its Gemini model with purported improvements in reasoning and factuality. Microsoft has expanded its AI Copilot offerings across its product suite. xAI has been testing its Grok-3 model on social media platforms, drawing both praise and criticism for its unfiltered responses. The absence of a transparent evaluation process raises concerns that these models may reach the public without adequate independent scrutiny.
Reactions from Experts
AI policy experts have expressed alarm at the page’s removal. “This is deeply concerning,” said a former NIST official who spoke on condition of anonymity. “The announcements were a key part of building public trust in the government’s AI safety efforts. Taking them down without explanation undermines that trust.” Another expert, a professor at a leading university specializing in AI governance, called for an immediate investigation and restoration of the page. “If the government is serious about transparency, it must provide clear reasons for any significant changes to public records.”
Some observers have speculated that the removal might be linked to internal disagreements within the administration about the proper balance between promoting AI innovation and imposing safety regulations. The America’s AI Action Plan had set ambitious goals for maintaining U.S. leadership, but some industry groups have argued that government-led evaluations create an unnecessary burden. The missing page could be a signal that the administration is reconsidering its approach, perhaps in response to lobbying from the companies involved.
However, there is no evidence to support any particular theory. The Commerce Department has not issued a statement, and the White House has remained silent. The redirect itself may have been implemented hastily, as the original URL now leads to the CAISI homepage without any intermediate notice. This suggests that the page was removed by a web administrator rather than through a planned content migration.
Technical Details of the Redirect
A check of the URL on Monday evening confirmed that it redirects to the CAISI main page at the Commerce Department’s NIST website using an HTTP 301 (Moved Permanently) status code. That means the old URL is permanently redirected. The Wayback Machine still holds copies of the original page, with the latest snapshot dated May 5, 2026. It is unclear whether the page will be restored or if a new URL will be provided. The rest of the CAISI website appears to be functioning normally, with other announcements, reports, and resources still available.
The incident highlights a broader issue with digital preservation in government. Many government agencies actively manage their websites, but content deletions are not always announced or archived. The Library of Congress’s Web Archive and the Wayback Machine provide valuable backups, but they cannot guarantee that every page is captured in a timely manner. If the page was removed within days of publication, there is a risk that some users may have missed it entirely.
Journalists and researchers have documented numerous examples of “link rot” on government sites, where URLs become inaccessible over time. A 2023 study by the Pew Research Center found that nearly 30% of links to U.S. government websites were broken. The vanishing of the CAISI page adds to that statistic and raises concerns about the reliability of government information.
What Comes Next
As the story continues to develop, attention is focused on the Commerce Department and the White House. If the page is restored, officials should provide an explanation for its removal and any changes made. If it remains taken down, Congress may seek answers during oversight hearings on AI policy. The issue could also become a talking point in the upcoming midterm elections, as candidates from both parties vie to demonstrate their commitment to government transparency.
For now, the only complete record of the announcement exists on the Wayback Machine and in news articles that reported on it. The question remains: why would a government body remove a page that seemed to showcase a successful collaboration with industry leaders to enhance AI safety? The answer may not be known until officials break their silence, if they ever do.
Source: Gizmodo News