Anthropic and Trump: Clash Over Fable Exportation

Meeting Between Anthropic and the Trump Administration

Dario Amodei, the CEO of Anthropic, recently met with officials from the Trump administration to discuss the export ban imposed on their artificial intelligence model, Fable. According to a report from POLITICO, this meeting aimed to resolve a dispute regarding perceived security vulnerabilities in the model.

Anthropic employees met on Monday with high-level officials from the Trump administration during their first in-person discussions after a federal government export ban forced the AI startup to withdraw its latest model from the market on Friday evening. Two administration officials, one familiar with the discussions and one close to the company, informed POLITICO, which is part of the Axel Springer Global Reporters network.

It is likely to take more than a few days to reach a resolution that alleviates the federal government's action from Friday, which prohibited Anthropic from allowing non-American users to access its latest model due to potential security vulnerabilities, said a senior White House official. However, this official left the door open to the possibility that this could be done quickly. "It's up to Anthropic to decide," he stated.

The company rushed to send senior executives with expertise in research and security to Washington after several hours of calls over the weekend between Anthropic co-founder Tom Brown, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross, officials said, who, like others quoted in this article, were granted anonymity to describe the discussions. Sarah Heck, Anthropic's public policy lead, was also present during these calls.

Monday's in-person meetings, organized by the Commerce Department and Cairncross's office, were more technical and staff-led, including Chris Fall, who heads the Commerce Department's Center for AI Standards and Innovation, the person familiar with the discussions clarified.

Anthropic presented its cybersecurity systems to administration officials in hopes of overcoming the restrictions, the administration official stated. Representatives from Anthropic included Logan Graham, who evaluates and tests models as part of the company's Frontier Red Team security team; Dave Orr, the company's security measures lead; and Nicholas Carlini, its principal security researcher, added a person close to Anthropic.

Implications for AI Innovation

This situation has raised concerns among artificial intelligence experts, who believe that such restrictions could hinder the United States' leadership in the field. The export controls imposed by the Trump administration are seen as a threat to American innovation and competitiveness against China.

The export restriction was the most significant escalation to date in the administration's attempts to regulate the rapidly evolving AI industry, which is outpacing the government infrastructure surrounding this revolutionary technology. It was also the first time the White House forced a company to withdraw a model from public access, but it was merely the latest in a series of conflicts between the administration and Anthropic regarding ethical restrictions on potentially catastrophic risks associated with AI.

Anthropic has defended its security measures since the restrictions were imposed on Friday evening. "These vulnerabilities all seem relatively simple, and we found that other publicly available models can also discover them without requiring a workaround," it stated in a blog post. The company also claimed that the security breach was minor and did not constitute a broader "jailbreak" allowing people to use it for cyberattacks.

Reactions and Perspectives

Treasury Secretary Scott Bessent, who played a key role in discussions with Anthropic CEO Dario Amodei before the administration imposed the export restrictions, was less involved over the weekend, and the Treasury did not participate in Monday's in-person meetings. Lutnick was solicited by the president and became more involved over the weekend due to his oversight of the country's export policy, the person familiar with the discussions stated. Both Lutnick and Bessent traveled on Air Force One with the president for G7 meetings taking place from Monday to Wednesday.

The export control, which prohibited Anthropic from allowing foreign nationals to use its latest model, Fable, effectively forced the company to completely withdraw the model to comply, the company stated on Friday evening.

The escalation occurred after a disagreement between Anthropic and the administration regarding a perceived security concern with Fable, which Amazon, a key investor in Anthropic, had raised with the White House over the weekend, as previously reported by POLITICO. Amazon reported finding a way to bypass Anthropic's protections on Fable, a highly anticipated model designed to be less powerful than the company's highly advanced Mythos model. Anthropic had allowed a select group of companies access to Mythos due to its ability to exploit cybersecurity vulnerabilities faster than any human.

Fable operated at a level similar to Mythos but had built-in protections to prevent users from pushing it to conduct a cyberattack or detail how to create a biological weapon, the company stated. Amazon's discovery, brought to the attention of the White House and the National Security Agency, prompted Cairncross, Bessent, and others to demand that Anthropic terminate access to Fable. When Anthropic did not comply, arguing instead that the security breach was minor and did not constitute a broader "jailbreak" allowing people to use it for cyberattacks, the administration imposed an export control that the company stated compelled it to act.

Over the weekend, a group of nearly 80 technical experts and CEOs wrote an open letter to the administration urging officials to revoke the export control. They stated that the actions that Amazon researchers had managed to execute with Fable constituted a "necessary capability in any model intended to write secure code" and "should not be considered an offensive capability."

The abrupt controls imposed by the Trump administration "have pulled the best models from defenders, created uncertainty in the market, and risked American leadership in AI without any real risk to justify it," the tech leaders wrote in the letter.

Graham, one of the Anthropic employees participating in the meetings with the administration, had told Fox News in April that Anthropic was seeking to find a balance by giving access to Mythos to defenders "without allowing malicious actors access." "Ultimately, we believe that tools like this need to be deployed securely worldwide, as we need the entire world to use these tools to defend themselves as well," Graham added. The interview took place before the release of Fable.

Some members of the tech community have supported the strict export restrictions and stated that the risks posed by breaching protections are serious. But other AI experts, and even officials within the administration, have stated that the controls create a threatening regulatory landscape that will stifle the industry.

"AI is now licensed, but the requirements are constantly changing and are always secret, even to the administration itself, which will discover the rules spontaneously in real-time in response to events," wrote Dean Ball, a former Trump administration official who worked on AI policy, on X.

A current administration official warned that the longer the export dispute lasts, the more likely it is to become a de facto licensing regime that AI companies will have to face when releasing new models, slowing their innovation and competition with China.

If the export control does not boil down to a simple "reprimand" quickly alleviated, "this will pose a huge problem for the entire industry," the administration official stated. "This means that every model in the future will have to ask the government for permission to know if it can be released. This is an extremely bad situation," the official added.