In a development that has sparked concern among transparency advocates and industry observers, a U.S. government webpage announcing agreements to vet unreleased artificial intelligence models from major tech companies has gone missing. The page, published by the Commerce Department’s Center for AI Standards and Innovation (CAISI), detailed partnerships with Google DeepMind, xAI, and Microsoft that would allow federal inspectors to conduct pre-deployment evaluations of frontier AI systems. As of late Monday, the original URL led to an error message before redirecting to the main CAISI page, effectively erasing the announcement from public view.
Background of the Agreements
The now-missing announcement, dated May 5, 2026, was a significant step in the Biden-Harris administration’s ongoing efforts to regulate the rapidly advancing field of artificial intelligence. It stated that CAISI, a division of the National Institute of Standards and Technology (NIST), had renegotiated and expanded previous partnerships with three of the most influential AI labs in the world. The agreements were designed to give the government insight into cutting-edge AI capabilities before those systems were released to the public, addressing concerns about potential risks ranging from bias and disinformation to catastrophic misuse.
Notably, the announcement built on earlier agreements signed in 2024 with Anthropic and OpenAI, which were among the first to grant the government pre-release access to their models. The inclusion of Google DeepMind, xAI, and Microsoft signaled a broadening of the government’s reach, encompassing companies that are developing some of the most powerful and widely used AI systems today. Google DeepMind is renowned for breakthroughs like AlphaFold and Gemini, while xAI, founded by Elon Musk, focuses on building “maximally curious” AI. Microsoft, a major investor in OpenAI, also develops its own AI models through Azure and partnerships with other labs.
The Disappearance and Initial Reactions
Reuters was the first to report the disappearance on Monday afternoon. Journalists attempting to access the original URL were greeted with the message “Sorry, we cannot find that page,” and later observed that the page automatically redirected to the main CAISI site. Attempts by Gizmodo to retrieve the announcement from the Wayback Machine succeeded: the archived version reveals the full text, including a quote from the original release: “These agreements support information-sharing, ensuring a clear understanding in government of AI capabilities and the state of international AI competition.”
As of Monday evening, no official explanation had been provided by either the White House or the Commerce Department. Gizmodo’s request for comment went unanswered, and the redirect remained in place. The timing of the removal is curious, coming just a week after the original announcement and amid heightened scrutiny of the administration’s AI policies. Critics argue that deleting such pages undermines public trust, especially when the agreements involve powerful private companies and potentially sensitive national security implications.
What the Agreements Entail
According to the archived version, the agreements are part of the government’s AI Action Plan, which focuses on testing frontier AI models for safety, security, and societal impact. The pre-deployment evaluations are intended to identify potential harms before the models are widely distributed, a process that has been voluntary but is increasingly seen as necessary given the pace of AI development. The evaluations are conducted by CAISI, which uses NIST’s AI Risk Management Framework as a baseline.
The original announcement also mentioned that the agreements had been “renegotiated to reflect CAISI’s directives from the secretary of commerce.” This suggests that the terms may have been updated to address evolving risks, such as the ability of AI to generate realistic deepfakes, automate cyberattacks, or manipulate financial markets. The deals reportedly allow government researchers to run their own tests on unreleased models—sometimes directly on the company’s internal servers—in what is known as pre-deployment red-teaming.
Wider Implications for AI Governance
The disappearance of the page raises broader questions about the transparency and durability of AI governance measures. If the administration can quietly remove announcements about key safety partnerships, how can the public be sure that other commitments are being upheld? This incident echoes earlier controversies where government websites were altered or deleted without notice, often leading to accusations of information suppression.
Moreover, the missing page comes at a time when the international community is racing to set rules for AI. The European Union has already passed the AI Act, while China has implemented strict licensing for generative AI. The United States has favored a mix of voluntary commitments and executive orders, but without consistent public messaging, the effectiveness of these efforts is diminished. The fact that a major announcement can vanish from the internet so quickly sends a troubling signal about the administration’s commitment to openness.
Industry experts have pointed out that the missing page might be the result of a simple website error or a routine content update, but the lack of a redirect or explanation suggests otherwise. If the page was intentionally removed, it could be because the agreements were revised or because the government does not want to draw attention to the extent of its involvement with these companies. Alternatively, it could be a technical glitch, but the fact that multiple news outlets confirmed the error and the redirect strongly indicates a deliberate action.
The Role of Key Stakeholders
Google DeepMind, xAI, and Microsoft have not issued public statements about the missing page. All three companies have previously expressed support for responsible AI development and have participated in various White House-led safety initiatives. However, critics argue that the real test lies in implementation: whether companies are genuinely allowing independent oversight or merely performing due diligence.
Anthropic and OpenAI, which signed similar agreements in 2024, have not commented on the situation. These two companies have been at the forefront of AI safety discourse, with Anthropic founding its business model on safety-first principles and OpenAI facing internal conflicts over the balance between safety and commercialization. Their early participation set a precedent, but the longevity of such agreements remains uncertain.
The Commerce Department’s silence is particularly concerning given that Secretary of Commerce Gina Raimondo has been a vocal advocate for AI regulation. In past speeches, she has emphasized the need for “guardrails” to prevent AI from causing harm, and CAISI was established in part to operationalize those guardrails. The disappearance of this page contradicts that message of transparency.
Historical Context of Government AI Announcements
This is not the first time a government AI-related announcement has vanished. In 2023, a White House fact sheet on AI safety commitments was briefly taken down before being restored with minor edits. Similarly, in 2024, a NIST document outlining AI risk assessment guidelines was removed for weeks without explanation. While these removals are often attributed to routine updates or corrections, they fuel skepticism among those who worry that the government is downplaying the risks of AI or bowing to corporate pressure.
The current incident is especially notable because it involves pre-deployment evaluations—a sensitive area where government access to proprietary technology could be seen as intrusive by some companies. It is possible that the agreements were never final, or that one party pulled out, leading to the removal of the announcement. However, without official confirmation, the public is left to speculate.
What Happens Next?
For now, the only record of the announcement exists in internet archives and in news articles that cited it. The Wayback Machine shows that the page was live for approximately one week, from May 5 to May 11, when it began redirecting. The archive captures the full text, including the quote about information-sharing and international competition. Gizmodo will update this article if a response from the White House or Commerce Department is received.
In the meantime, the incident serves as a reminder that government websites are not permanent records, and that public transparency requires continuous effort. The AI community and the public at large are left to wonder whether the agreements are still in effect, and if so, how they will be enforced without a clear public-facing endorsement. As AI models grow more powerful, the need for clear, stable, and accessible governance mechanisms becomes ever more critical.
The disappearance of the page also highlights the importance of independent organizations that track changes to government content. Groups like the Sunlight Foundation and the Electronic Frontier Foundation have long advocated for greater transparency in government digital communications. Without their vigilance, actions like this might go unnoticed. The fact that the Wayback Machine preserved the announcement underscores the value of archival efforts.
This story is still developing, and more details may emerge in the coming days. It is possible that the page will reappear with updated language, or that an official explanation will be provided. Until then, the missing announcement stands as a cautionary tale about the fragility of digital government records and the importance of holding public institutions accountable for their communications. The underlying agreements, if they remain active, represent a milestone in AI governance, but their credibility is now partly undermined by the secrecy surrounding their public documentation.
Source: Gizmodo News