Just over a week ago, the Department of Commerce’s Center for AI Standards and Innovation (CAISI) made a significant announcement: it had reached agreements with three major AI companies—Google DeepMind, Microsoft, and xAI—to allow the government to inspect unreleased AI models before public deployment. The move was framed as a critical step toward ensuring frontier AI capabilities are safe, secure, and aligned with national interests. However, that announcement has now vanished from the CAISI website. The original URL, which once contained a detailed press release dated May 5, 2026, now leads to an error page or redirects visitors to the main CAISI landing page. The disappearance was first reported by Reuters, and as of late Monday evening, the missing page has not been restored.
The sudden removal of the page raises immediate questions about government transparency. The agreements themselves were described as expanded industry collaborations that would enable CAISI to conduct pre-deployment evaluations and targeted research. An archived version of the announcement, preserved by the Wayback Machine, states: “Today, the Center for AI Standards and Innovation (CAISI) at the Department of Commerce’s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.” The archived text goes on to mention that these agreements build on previously announced partnerships, renegotiated to reflect directives from the secretary of commerce and America’s AI Action Plan.
This type of pre-deployment vetting is not entirely new. In 2024, Anthropic and OpenAI signed similar agreements with the government, setting a precedent for voluntary information-sharing and safety testing. Those earlier pacts were hailed as a model for responsible AI governance, especially as concerns grew about the potential for advanced AI systems to be misused or to cause unintended harm. The inclusion of Google DeepMind, Microsoft, and xAI in 2026 was seen as expanding the scope of oversight to cover more of the influential players in the AI landscape. xAI, founded by Elon Musk, has been particularly vocal about AI safety, making its participation notable.
Why the page was removed remains unclear. The Commerce Department and the White House did not immediately respond to requests for comment. It is possible the removal is due to a technical error, a routine website update, or a deliberate decision to recalibrate the messaging around these agreements. However, the lack of an explanation fuels speculation. Some observers point to the broader political environment surrounding AI regulation. The Biden administration had pursued a mix of voluntary commitments and executive orders; the Trump administration, which began in January 2025, has signaled a different approach. The mention of “America’s AI Action Plan” in the archived text suggests the current administration aims to assert its own vision for AI governance.
The missing page also highlights the fragility of digital government records. When a URL disappears without a redirect or an archived copy, the information can become inaccessible to journalists, researchers, and the public. The Wayback Machine’s archived version of the announcement includes additional details that were not widely reported: the agreements support “information-sharing” and “ensuring a clear understanding in government of AI capabilities and the state of international AI competition.” This language underscores that the pacts are not just about safety but also about maintaining a competitive edge over foreign adversaries, particularly China, which has invested heavily in AI development.
The agreements themselves involve access to frontier AI models—those near the cutting edge of capability. For example, Google DeepMind’s Gemini models, Microsoft’s in-house and partner models, and xAI’s Grok series are all likely candidates for evaluation. The government’s ability to test these models before public release could help identify vulnerabilities, biases, or potential for misuse. In a field where new capabilities emerge rapidly, such oversight could be indispensable. However, the voluntary nature of these agreements means companies can withdraw at any time, and the lack of a public-facing announcement may erode trust in the process.
Historically, government-industry collaborations on emerging technologies have faced similar transparency challenges. During the early days of nuclear energy, details of safety agreements were often classified. More recently, cybersecurity information-sharing initiatives have struggled to balance openness with protection of proprietary data. The AI agreements walk a similar tightrope: companies want to protect trade secrets, while the government aims to assure the public that systems are being scrutinized. The removal of the page, whether intentional or accidental, undermines that assurance.
The implications extend beyond transparency. If the page was removed because the administration is reconsidering the agreements, that could signal a shift in AI policy. Alternatively, if it is a technical glitch, the delay in restoration suggests a lack of urgency. Either way, the episode reveals the precarious nature of AI governance in the United States. Unlike the European Union’s AI Act, which imposes binding regulations, the U.S. relies heavily on voluntary commitments and executive action. When a simple webpage goes missing, the entire edifice appears shaky.
For the companies involved, the removal may be inconsequential; their internal processes continue. But for the public, it reinforces the sense that AI oversight is opaque and unreliable. The archived announcement also mentions “renegotiated” partnerships, implying that earlier agreements had been modified. What those modifications entailed is not public. And with the page gone, the opportunity for public scrutiny has evaporated.
As of Monday night, Gizmodo had not received a response from either the White House or the Commerce Department. The story, however, is likely to persist. Journalists and watchdogs will continue to press for answers. The Wayback Machine’s snapshot ensures the original text is not lost, but the official silence speaks volumes. The episode serves as a reminder that in the age of digital government, a missing page can be as telling as a written policy.
Source: Gizmodo News