Anthropic staff held their first face-to-face meeting with Trump administration officials on Monday, two government officials, a person familiar with the discussions and a person close to the company told Politico after a federal export ban forced the artificial intelligence startup to pull its latest model from the market Friday night.
A White House official said it would likely take several days or more to pass a resolution relaxing the federal government’s Friday action against Anthropic, which barred users outside the United States from accessing its latest models due to potential security vulnerabilities. But officials left the door open to the possibility that it could happen soon.
“That’s up to Anthropic,” the official said.
After an hours-long phone conversation over the weekend between Anthropic co-founder Tom Brown, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross, the company rushed to send senior executives with investigative and security expertise to Washington, D.C., said the officials, who spoke on condition of anonymity to describe the discussions, as did others cited in this article. Sarah Heck, Director of Humanity for Public Policy, was also present on the call.
Monday’s in-person meeting was held by Commerce’s Cairncross offices and was more technical in nature, led by staff including Chris Fall, director of Commerce’s Center for AI Standards and Innovation, said a person familiar with the discussions.
Anthropic made a presentation to government officials explaining Anthropic’s cybersecurity safeguards in hopes of overcoming the restrictions, government officials said.
Representatives from Anthropic included Logan Graham, who evaluates and stress tests models as part of the company’s Frontier Red team; Dave Orr, the company’s safety officer. According to a source close to Anthropic, Nicholas Carlini, the company’s head of security research, said:
The export restrictions are the most significant escalation yet in the administration’s attempts to regulate the AI industry, which is developing more rapidly than the government’s infrastructure, which is centered around breakthrough technology. It’s also the first time the White House has forced a company to remove a model from public access, but it’s just the latest in a series of arguments between the administration and Anthropic over ethical limits on potentially catastrophic AI risks.
Anthropic has been adhering to its safety measures since the restriction order went into effect Friday night.
“While both of these vulnerabilities appear relatively simple, we have found that other publicly available models can similarly discover these vulnerabilities without the need for bypass,” the company said in a blog post.
Treasury Secretary Scott Bessent played a leading role in discussions with Anthropic CEO Dario Amodei before the administration instituted export controls, but his involvement waned over the weekend and Treasury did not participate in Monday’s in-person meeting.
Mr. Lutnick, who was appointed by the president to oversee the country’s export policy, became more involved over the weekend, according to people familiar with the discussions. Lutnick and Bessent flew with the president to France on Air Force One for the G7 meeting, which runs from Monday to Wednesday.
Export regulations that prohibit Anthropic’s latest model, the Fable 5, from being used by foreigners have effectively forced the company to completely discontinue the model in order to comply, the company announced Friday night.
As POLITICO previously reported, the escalation came after a disagreement between Anthropic and the administration over security concerns about Fable that Amazon, a major investor in Anthropic, raised with the White House late last week.
Amazon has reported that it has found a way around Anthropic’s guardrails with Fable, a long-awaited model designed to be less powerful than the company’s advanced model Mythos. Anthropic has only allowed access to Mythos to a select group of companies because of their ability to exploit cyber vulnerabilities faster than humans.
Fable operated at a similar level to Mythos, but had built-in safeguards to prevent users from prompting details of how to carry out cyberattacks or create biological weapons, the company said. The Amazon discovery escalated to the White House and National Security Agency, where Cairncross, Bessent and others demanded that Anthropic cut off access to Fable.
When Anthropic disagreed, insisting that the security breach was minor and not a large-scale so-called “jailbreak” that people could use to launch cyberattacks, the administration imposed export controls, which the company claimed were enforced.
Last weekend, a group of nearly 80 technology experts and CEOs wrote an open letter to the government, calling on authorities to lift export restrictions. They said the actions Amazon researchers asked Fable to perform were “necessary features in a model intended to write secure code” and “should not be considered offensive features.”
The Trump administration’s surprise regulations “deprived defenders of the best models, created market uncertainty, and jeopardized America’s AI leadership with no real risk to justify it,” the tech leaders said in the letter.
Graham, one of the Anthropic employees who attends meetings with the administration, told Fox News in April that Anthropic is walking a fine line in giving defenses access to Mythos “without giving the bad guys access.”
“Ultimately, we think we need to extend tools like this to the entire world in a secure way, because the entire world also needs to use these tools to protect themselves,” Graham added. The interview was conducted prior to the release of Fable.
Some in the technology industry support stricter export controls, saying the risks posed by breaching the guardrails are serious. But other AI experts and even administration officials say the controls create a threatening regulatory environment that will hold the industry back.
“Although AI is currently authorized, the requirements are constantly changing and will always remain secret, even to the administration itself. The government will spontaneously discover rules in real time as it reacts to events,” Dean Ball, a former Trump administration official who worked on AI policy, wrote about X.
One current administration official warned that the longer the export dispute drags on, the more likely it will become a de facto licensing system that AI companies will have to navigate when launching new models, slowing innovation and competition with China.
If export restrictions are not eased soon with a short “slap on the wrist,” “this will be a huge problem for the entire industry,” the official said.
“This means that all future models will have to get government approval for launch. This is a very bad situation,” the official said.
Megan Messerly contributed to this report.
This story was originally Appeared on POLITICO Courtesy of the Axel Springer Global Reporters Network, which leverages the resources of its newsroom to publish ambitious scoops, investigations, interviews, opinion pieces and analysis. This allows journalists from POLITICO, Business Insider, WELT, BILD, Onet, Fakt and more to collaborate on important stories across platforms and to an international audience of hundreds of millions of people.