Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Trump administration officials ended negotiations with Anthropic on Monday without removing existing controls was installed last week more in the company examples of advanced AI in response to prison concerns, according to three people briefed on the matter.
Administrators continue to believe there are ways to block some of the content on Anthropic’s Claude Fable 5, allowing users to take advantage of the company’s Mythos cybersecurity capabilities, the people said.
Anthropic has said for days that the administration’s concerns are increasing, a position that he reiterated during working group meetings held at the Commerce Department with government researchers from the Center for AI Standards and Innovation (CAISI) and the Office of National Cyber Director Sean Cairncross, one of the people said.
The meetings were also attended by Commerce Secretary Howard Lutnick, who made the call during the G7 summit in Evian, France. Cairncross himself was not involved, the person said.
On the Anthropic side, cofounder and chief computing officer Tom Brown and director of external affairs Sarah Heck will be leading the discussion. Anthropic’s red team leader, Logan Graham, and senior security researcher Nicholas Carlini flew to Washington, DC for talks.
“Both teams are working hard to make this possible,” 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 would be. The Commerce Department has expressed interest in finding a way to bring Fable 5 online for consumer use, but it may be up to Anthropic to address jailbreak concerns, the person said.
Calling the Alarm
The emergency talks have come at a critical time for Anthropic politics, which was already long overdue confronting the Pentagon if its AI models can be used for other military purposes.
The Trump administration was first informed of the prison concerns last week. Amazon CEO Andy Jassy called Treasury Secretary Scott Bessent directly about the perceived threat, sparking a flurry of leadership, the people said. Jassy’s conversation with the Trump administration was the first report by The Information.
Alarmed White House officials tasked the NSA with helping to reassess the problem. The NSA responded that it believed it was indeed possible to strip the Fable 5’s surveillance system, prompting the administration to place restrictions on the model.
Lutnick then spoke with the director of Anthropic Dario Amodei on Friday, when the Department of Commerce wrote its letter to grant foreign authorities of Fable 5. At the end of the week, after Anthropic cut off access to the model for all users, Lutnick was on several calls with Brown and Heck, according to a person with knowledge of the situation.
It’s not clear why Amazon, one of Anthropic’s biggest investors, raised the alarm over Fable 5. “As a leading cloud provider that serves many private and government customers, it’s not uncommon for governments to seek our advice on potential security threats,” an Amazon spokesperson tells WIRED. “When it happens, we will not share the details of this conversation.”
Separation Protection
At the heart of the conversation between Anthropic and the leadership is a disagreement over the difficulty of Claude Fable 5’s prison break.
In a blog post On Friday, Anthropic reported that the leadership’s view on potential threats is overwhelming. Some cybersecurity researchers reiterated this to officials on Monday, posting an open letter they argue that Anthropic’s actions in controlling imports were unfair.
“Anthropic’s Mythos-class is very good at finding bugs and weapons. However, it is not particularly good at this task, and many of its signatories regularly use other basic and visible examples in security research and the red team every day,” the open letter reads. “As a result, this has removed the best examples for defenders, created market uncertainty, and jeopardized America’s AI leadership without risk of justification.”
