Trump administration officials ended talks with Anthropic on Monday, according to three people briefed on the matter, without lifting export controls imposed last week on the company’s most advanced artificial intelligence models in response to jailbreaking concerns.
The administration still believes there are ways to disable some of Anthropic’s Claude Fable 5 defenses, effectively allowing users to access the more powerful cybersecurity features of the company’s Mythos model, the people said.
Anthropic has argued for days that the administration’s concerns are overblown, a point it reiterates during working group meetings held at the Commerce Department with government researchers at the Center for AI Standards and Innovation and the Office of the National Cybersecurity Director, Sean Cairncross, one of the people said.
Secretary of Commerce Howard Lutnick also attended the meetings and reached the G7 summit in Evian, France by telephone. Cairncross himself was not involved, the person added.
On Anthropic’s side, the discussions are led by co-founder and chief computation officer Tom Brown and head of external affairs Sarah Heck. Anthropic’s border red team leader, Logan Graham, and senior security researcher Nicholas Carlini flew to Washington for talks.
“Both parties are working quickly to resolve this issue,” an Anthropic spokesperson said in a statement to WIRED. A White House spokesman declined to comment on the matter.
It was not immediately clear what the next steps would be. The Commerce Department has expressed interest in finding a way to restore Fable 5 for online consumer operate, but that will likely be contingent on Anthropic fully resolving the jailbreak issues, the person said.
Ringing the alarm
The emergency talks came at a tense political moment for Anthropic, which was already in a long-running dispute with the Pentagon over whether its artificial intelligence models could be used for certain military applications.
The Trump administration was first alerted to concerns about the prison break last week. Amazon CEO Andy Jassy called Treasury Secretary Scott Bessent directly about alleged security vulnerabilities that played a role in spooking the administration, the people say. The first was Jassy’s conversation with the Trump administration reported via Information.
Concerned White House officials tasked the NSA with helping check the vulnerabilities. The NSA responded that it believed it was indeed possible to remove the guardrail from Fable 5, prompting the administration to impose restrictions on the model.
Lutnick then spoke with Anthropic CEO Dario Amodei on Friday while the Commerce Department was preparing a letter imposing export controls on Fable 5. Over the weekend, after Anthropic cut off all users’ access to the model, Lutnick spoke with Brown and Heck multiple times, according to a person familiar with the events.
It’s unclear why Amazon, one of Anthropic’s largest investors, raised the alarm on Fable 5. “As a leading cloud services provider that serves a large number of private and public sector customers, it is not uncommon for governments to seek our advice on potential security risks,” an Amazon spokesperson tells WIRED. “When they happen, we do not share the details of those discussions.”
Security disconnection
At the heart of the talks between Anthropic and the administration is a disagreement over the severity of concerns related to the Claude Fable 5 jailbreak.
In blog post on Friday, Anthropic suggested that the administration’s characterization of potential threats is exaggerated. Some cybersecurity researchers reiterated that stance to officials on Monday in a message open letter arguing that the export control actions taken against Anthropic were unjustified.
“Anthropic’s Mythos models are quite good at finding flaws and weaponizing exploits. However, they are not exceptionally good at these tasks, and many of the undersigned regularly use other base models and open source models on a daily basis for security audits and red-teaming,” the open letter reads. “As a result, this action took away defenders’ best models, created market uncertainty, and threatened America’s AI leadership, without any real risk to justify it.”