Confusion Reigns Over White House’s AI Whitelist

It may be one of the hottest tickets in corporate America right now: membership in the “trusted partner” program that allows companies to get access to the most advanced AI models before they’re released to the public. There’s just one problem: no one outside of the Trump administration has a clear idea how to get on that list.

As laid out in the administration’s executive order in early June, which established a voluntary testing framework for frontier AI models, these “trusted partners”— including banks, cybersecurity firms and power utilities—would use their access to sniff out potential vulnerabilities in their computer systems exposed by the new AI models.

But the White House has created a de facto “whitelist” of companies allowed on the list, according to multiple people familiar with the process, including an executive from a top Wall Street firm. The executive added that there was a worrying lack of clarity over how organizations can be included, which they said was frustrating given that “we have this opportunity to proactively mobilize a set of solutions here and not wait for a crisis moment to get that underway.”

The trusted partner program is central to governmental and corporate efforts to shore up critical infrastructure for the AI era. Labs including OpenAI and Anthropic keep raising the alarm about the capabilities of new models and the need for a government-coordinated response to the threats that accelerating AI development poses. Uncertainty about membership in the trusted partner program highlights how the Trump administration’s AI agenda has been defined by opaque and often contradictory decisions.

A White House spokesperson did not respond to a request for comment.

One security executive at a top utility company said that they did not receive access to Anthropic’s Mythos for months after its release. Anthropic told them the White House was involved in approving access but the utility couldn’t figure out whether they were being blocked by the government or Anthropic. They similarly didn’t get access to OpenAI’s most cyber-capable model that it released to a select group in August; they’re set to finally receive access to both labs’ models in the fall.

The government’s “trusted partner” program has effectively absorbed programs originally set up by Anthropic and OpenAI to open their most advanced models to certain companies for cybersecurity testing. Anthropic launched Glasswing for that purpose in April and OpenAI has a similar program called Daybreak. Neither company has said exactly how they determine who can participate in those programs.

And now the White House appears to be involved. Multiple companies that are part of OpenAI and Anthropic’s early access programs that spoke with The Information received the impression that final approval for their involvement was ultimately up to the White House’s Office of the National Cyber Director.

Back and Forth

One person familiar with the leading AI labs said they have not seen a White House master list of companies allowed early access, describing the trusted partner program instead as a back-and-forth between the labs and government.

Scrutiny over new models, and the push for additional cyber defenses, has grown due to recent incidents like the Hugging Face attack, where OpenAI agents broke out of a testing sandbox to hack into the open source repository.

White House officials have repeatedly stressed that all elements of the framework, including the trusted partner program, are voluntary. But the White House has grown increasingly involved in the rollout of new models since the announcement of Anthropic’s Mythos in April triggered broader concerns within the administration about the cyber capabilities of models.

The White House imposed export controls on Fable, the public version of Mythos, in June after concerns were raised about the model making it easier for hackers to breach software protections. The same month, OpenAI CEO Sam Altman informed staff that the government would be approving access “customer by customer” for the preview period of its latest model, GPT 5.6, The Information previously reported.

Industry executives had hoped that the finalization of the White House’s voluntary frontier model testing framework, which was to be released on Aug. 1, would create more concrete guidance, putting more control back in the hands of AI labs and creating clear guardrails around the government’s role.

In the Dark

After briefing a select group of companies on the framework in early August, the White House has yet to release a public version, leaving most of the industry in the dark. That has left unanswered questions over whether open models will be subject to the framework in the future, as well as how the trusted partner program will function.

One company representative said they scheduled meetings with the Office of the National Cyber Director and the White House’s Office of Science and Technology Policy in recent weeks to push to get on the government’s “whitelist” for early access, as the representative described it. They found out the company was already included, illustrating continued confusion between the labs and government over the process.

The company representative added that both White House offices communicated that the clearance and early access process could apply to open-source models in the future, though they did not provide a timeline. Currently, the framework only applies to closed frontier models from leading labs like OpenAI and Anthropic. It is unclear how the framework would apply to open-source models, especially from Chinese developers such as Moonshot and DeepSeek, as well as how companies would get early access to the models before public release.

Even as the labs and White House continue to develop the process, the lack of concrete standards could create further complications. The Wall Street executive said that more clarity about the access approval process and its links to other public cybersecurity initiatives would support better collaboration between private industry and the government.

For example, the June executive order also established Gold Eagle, a so-called clearinghouse where government agencies and companies could coordinate on cyber vulnerabilities and responses. But those programs are less effective, the person said, if there isn’t clear communication from the White House about the role of companies getting early access to frontier models.

Aaron Holmes, Stephanie Palazzolo and Laura Bratton contributed to this article.

Leo Schwartz is a reporter covering the intersection of tech and politics. He can be reached at [email protected] or on Signal at leomschwartz.24