Anthropic is negotiating with the US government to resolve export restrictions on its AI models.

Anthropic is seeking to resolve a new conflict with the Trump administration over the issue of artificial intelligence security, which has led the company to shut down global access to its two most advanced AI models. According to a source familiar with the matter, senior technical staff from Anthropic met with Department of Commerce officials on Monday to discuss the government's national security concerns. It is not clear which government officials attended the meeting. Following the meeting, a spokesperson for Anthropic said that both sides are working quickly to find a solution. A spokesperson for the Department of Commerce has not commented on the matter. Discussions began shortly after the Trump administration requested that Anthropic stop providing its cutting-edge products Fable 5 and Mythos 5 to foreign citizens. Since then, representatives from Anthropic have held multiple online meetings with US officials to discuss specific security concerns. On Friday afternoon, the Department of Commerce issued a directive to Anthropic, marking the most significant government intervention in the operations of an AI company to date. Just weeks ago, Anthropic submitted a confidential initial public offering (IPO) application with a valuation exceeding $90 billion. This directive presents a new challenge to the company. On Friday evening, Anthropic announced on its website that it had complied with the US government's directive to disable access to both models. The company stated that they believed the US government issued the order after discovering that the security measures for Fable 5, which Anthropic had previously prevented from performing cybersecurity tasks, could be "compromised". Fable 5 is the latest version of the Mythos series. In a statement, Anthropic said, "We do not believe that simply discovering a potential edge-case vulnerability necessitates a recall of a commercially-deployed model serving billions of users. Applying this standard across the industry would effectively block all providers of cutting-edge models from deploying new models." Privately-owned Anthropic has positioned itself as a more responsible AI developer, sharing its Mythos model with a very limited number of companies and institutions for the first time in April and warning of potential cybersecurity risks associated with wider distribution. Last week, the company released Fable 5, the first public version of its Mythos model, with protective measures in place to restrict its full network capabilities. Ayham Boucher, Executive Director of AI Strategy and Innovation at Cornell University, described the debate as simply requesting Anthropic to "fix the fence" and then lift the ban on Fable 5, but he cautioned that this underestimates the difficulty of addressing a highly technical problem. The disclosure of the AI vulnerability by Anthropic has accelerated the Trump administration's efforts in AI policy, leading to an executive order signed earlier this month requiring voluntary reviews of advanced AI models. The order specifies that none of its provisions should be construed as establishing mandatory licensing requirements. The Department of Commerce instructed Anthropic on Friday to stop offering its products to foreign citizens, bringing new uncertainty for companies and allied governments seeking to comply with government policy. "Clearly, the news over the past few days has been a big wake-up call for the industry as a whole," said Joelle Pineau, Chief AI Officer at Canadian AI developer Cohere Inc. "Just a few days ago considered one of the strongest models in the industry, now suddenly unavailable, this suggests there may be significant volatility in technology acquisition," Pineau said in an interview on Monday. "This is certainly a major concern for businesses." This dispute marks the latest conflict between the US government and the company. For months, Anthropic has been at odds with the Pentagon over additional restrictions sought for its AI tools in military applications. After contract negotiations fell through, the Department of Defense announced in March that the company posed a supply chain risk and was seeking other AI vendors to serve the military.