Trump administration officials He concluded talks with Anthropic on Monday without lifting export controls imposed last week on the company’s most advanced AI models in response to prison concerns, according to three people briefed on the matter.
Administrators continue to believe there are ways to disable some of Anthropic’s Claude Fable 5 security measures, effectively allowing users to access the powerful cyber security capabilities of the company’s Mythos model, the people said.
Anthropic has said for days that management concerns are overblown, a position it expressed in working group meetings held at the Commerce Department with government researchers from the Center for AI Standards and Innovation and the Office of the National Director of Cyber, Sean Cairncross, one of the people said.
The meetings were attended by Commerce Secretary Howard Lutnick, who called from the G7 summit in Evian, France. Cairncross himself was not involved, the person said.
On the Anthropic side, founder and chief computing officer Tom Brown and head of external affairs Sarah Heck were leading the discussions. The head of Anthropic’s red team, Logan Graham, and senior security researcher Nicholas Carlini flew to Washington, DC, for talks.
“Both parties are working quickly to resolve this,” an Anthropic spokesperson said in a statement to WIRED. A White House spokesman declined to comment.
It was not immediately clear what the next steps might be. The Commerce Department has expressed willingness to find a way to bring Fable 5 back online for consumer use, but that may depend on Anthropic fully addressing the jail’s concerns, the person said.
The Alarm is going off
The emergency talks come at a politically charged time for Anthropic, which has been embroiled in a long battle with the Pentagon over whether its AI models can be used for certain military applications.
The Trump administration was first made aware of the prison’s concerns last week. Amazon CEO Andy Jassy called Treasury Secretary Scott Bessent directly about the alleged risk, which played a role in angering executives, the people said. Jassy’s interview with the Trump administration was first reported by The Information.
Alarmed White House officials tasked the NSA with helping review the vulnerabilities. The NSA responded that it believed it was indeed possible to hack the Fable 5’s surveillance cables, prompting the administration to put restrictions on the model.
Lutnick then spoke with Anthropic CEO Dario Amodei on Friday, as the Commerce Department drew up its letter imposing export controls on the Fable 5. Over the weekend, after Anthropic cut off access to the model for all users, Lutnick had several phone calls with Brown and Heck, according to a person with knowledge of the incident.
It’s unclear why Amazon, one of Anthropic’s biggest investors, sounded the alarm on Fable 5. “As a leading cloud provider serving a large number of private and public customers, it’s not uncommon for governments to seek our advice on potential security risks,” an Amazon spokesperson told WIRED. “In the event, we do not provide details of these discussions.”
Security Termination
At the heart of the discussions between Anthropic and management is a disagreement about the extent of Claude Fable 5’s jailbreak problems.
In a blog post on Friday, Anthropic pointed out that management’s characterization of potential risks is overblown. Some cybersecurity researchers reiterated this position to officials on Monday, sending an open letter saying the export control action taken against Anthropic was unfair.
“Anthropic’s Mythos class models are very good at detecting bugs and exploits. However, they are not particularly good at these tasks, and many of the undersigned often use other basic and open source models for security testing and red team integration on a daily basis,” the open letter reads. “As a result, this action has removed the best models from defenders, created market uncertainty, and jeopardized America’s AI leadership without any real risk to justify it.”
