In a move that has raised eyebrows among transparency advocates and industry watchers, the U.S. Commerce Department’s Center for AI Standards and Innovation (CAISI) has quietly removed a webpage that announced pioneering agreements with three of the world’s leading artificial intelligence companies: Google DeepMind, Microsoft, and xAI. The page, which was published on May 5, 2026, detailed plans for the government to conduct pre-deployment evaluations of frontier AI models before they are released to the public. As of May 11, 2026, accessing the original URL leads to a redirection to the main CAISI page on the Department of Commerce website.
The deleted announcement had stated that CAISI would “conduct pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security.” It further explained that these agreements built on previously announced partnerships with OpenAI and Anthropic, which were signed back in 2024, and had been renegotiated to align with directives from the secretary of commerce and America’s AI Action Plan. The page emphasized that the collaborations “support information-sharing” and help ensure “a clear understanding in government of AI capabilities and the state of international AI competition.”
Background on CAISI and Pre-Deployment Evaluations
CAISI was established within the National Institute of Standards and Technology (NIST) to lead the development of AI standards and to foster innovation in AI safety. A key part of its mission is to work with industry to evaluate cutting-edge AI systems before they are widely deployed. The concept of pre-deployment evaluations is not new; it mirrors practices in other high-stakes industries such as pharmaceuticals and aviation, where new products are rigorously tested before reaching consumers. In the AI domain, such evaluations are intended to identify potential risks such as bias, security vulnerabilities, or the capability to generate harmful content, and to ensure that companies are adhering to voluntary safety commitments.
The original agreements with OpenAI and Anthropic in 2024 were seen as a landmark step toward voluntary government-industry collaboration on AI safety. Those deals allowed CAISI researchers to access unreleased models to conduct assessments. The new agreements with Google DeepMind, Microsoft, and xAI were supposed to extend this framework to a broader set of frontier AI developers. Microsoft, through its deep investment in OpenAI, and xAI, founded by Elon Musk, are major players in the generative AI space.
The Disappearance of the Webpage
The removal of the announcement was first noticed by journalists who attempted to access the original link on May 11, 2026. Instead of the expected content, the URL initially returned a “Sorry, we cannot find that page” error. Shortly thereafter, the Commerce Department configured the link to redirect to the main CAISI page. The Wayback Machine, an internet archive, preserved a copy of the original announcement, allowing the public to view what had been stated.
The Commerce Department has not issued a formal explanation for the removal. When contacted for comment, representatives did not immediately respond. However, such actions often occur when governments reevaluate the content or timing of public announcements. It is possible that the page was taken down to be revised, or that the details of the agreements are being updated. Alternatively, the removal could be a response to concerns that the announcement oversold the scope or impact of the deals, or that it contained sensitive information that should not be public.
Given the political climate surrounding AI regulation, the disappearance is particularly notable. The U.S. government has been under pressure from both industry and consumer advocates to clarify its approach to AI oversight. Some critics argue that voluntary agreements are insufficient and that mandatory regulation is needed. Others worry that too much transparency could hamper innovation or reveal national security vulnerabilities.
Broader Implications for AI Oversight
This incident highlights the opacity that can surround government-technology industry partnerships. Even when deals are announced with fanfare, the actual terms, evaluation methodologies, and results may remain undisclosed. The removal of the CAISI page could be an isolated administrative action, but it could also signal a shift in strategy. If the government is reconsidering the terms of its engagement with AI companies, it may choose to release a more detailed or modified announcement later.
The timing is also critical. The AI Action Plan mentioned in the deleted announcement is a comprehensive framework that the Biden administration unveiled to promote responsible AI development. It includes provisions for safety testing, workforce training, and international cooperation. The CAISI agreements were a tangible step in implementing that plan. Their sudden withdrawal from the public eye could undermine confidence in the government’s commitment to transparency.
Moreover, the three companies involved—Google DeepMind, Microsoft, and xAI—are formidable forces in AI. Google DeepMind is a leader in foundational research, Microsoft has integrated AI into its entire product ecosystem, and xAI, though newer, has attracted significant attention with its Grok model. The extent to which they will continue to cooperate with government evaluations is unclear. While the agreements are legally binding, their enforcement may be limited if the public cannot scrutinize compliance.
Historical Context: Pre-Deployment Testing in Other Industries
The concept of pre-market testing is well established in sectors such as automotive safety, where the National Highway Traffic Safety Administration (NHTSA) conducts crash tests before vehicles can be sold. Similarly, the Food and Drug Administration (FDA) requires clinical trials for new drugs. In the technology realm, however, such testing has historically been less formal. AI models are often released without independent verification of their capabilities or risks. The CAISI initiative was an attempt to create a new norm of voluntary pre-deployment testing.
Critics have noted that voluntary agreements lack teeth compared to mandatory regulations. They point to previous incidents where AI systems exhibited harmful behavior after release, such as generating biased outputs or spreading misinformation. The removal of the CAISI page only adds to concerns that the government is not fully committed to robust oversight. Supporters, on the other hand, argue that voluntary collaboration can be more agile than rulemaking processes, allowing standards to evolve quickly with the technology.
What Happens Next?
At the time of writing, the Commerce Department has not announced any plans to restore the page or issue a replacement. The companies involved have also remained silent. It is possible that the agreements are still in effect and that the page removal is a technical or administrative error. However, given the sensitivity of AI policy, it is more likely that stakeholders will demand answers. Media inquiries, congressional oversight, and public interest groups may pressure the department to clarify its intentions.
The incident serves as a reminder that even publicly announced government initiatives can vanish from the digital record without explanation. For those following AI governance, this event underscores the importance of archival efforts and constant vigilance. The full text of the original announcement, preserved by the Wayback Machine, continues to circulate among reporters and analysts.
In the coming weeks, we may see a revised version of the announcement appear, perhaps with updated terms or a more detailed description of evaluation procedures. Alternately, the government may decide to handle such agreements confidentially, which would further reduce public insight into how frontier AI is being assessed. Either way, the removal has already sparked debate about the balance between secrecy and transparency in the age of advanced AI.
Source: Gizmodo News