White House and Anthropic fail to end export ban on Claude Fable 5 AI model
Negotiations between Anthropic and the Trump administration over restrictions on the Claude Fable 5 model ended Monday with no agreement, leaving the first-ever forced removal of an AI model in place.
Background tension
The White House had already lost trust in Anthropic weeks before the export ban was imposed, according to unnamed officials cited by the Washington Post. The rift centred on ethical restrictions the company maintained around potentially catastrophic AI risks, with administration officials beginning to weigh sanctions well before Friday’s directive.
The jailbreak threat
Concerns focus on vulnerabilities that could allow users to strip the guardrails from Claude Fable 5 and access the more powerful cybersecurity capabilities of Anthropic’s internal Mythos model. Amazon CEO Andy Jassy called Treasury Secretary Scott Bessent last week to flag the issue, which prompted the NSA to review the matter. The agency concluded the jailbreak was indeed possible, according to people familiar with the discussions.
Weekend diplomacy
After the ban was imposed on Friday night, senior figures held hours‑long calls over the weekend. On Anthropic’s side were co‑founder Tom Brown and head of policy Sarah Heck; from the administration, Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross participated. The company then raced to send technical staff to Washington.
Both parties are working quickly to get this resolved.
Monday’s deadlock
In‑person meetings at the Commerce Department included detailed presentations by Anthropic’s frontier red team and safeguards experts Logan Graham, Dave Orr and Nicholas Carlini. The administration was represented by Chris Fall, who heads the Center for AI Standards and Innovation. Despite the technical exchanges, the talks ended with export controls still in place. A senior White House official said it would likely take longer than a few days to reach a resolution, though the door was left open to a faster outcome.
It will likely take longer than a few days to reach a resolution.
Wider context
The restriction marks the most significant escalation in the US government’s attempts to regulate the quickly advancing AI industry. It is the first time the White House has forced a company to remove an AI model from public access. For Anthropic, the clash comes amid a separate fight with the Pentagon over military use of its models.
These vulnerabilities all appear relatively simple, and we have found that other publicly‑available models are able to discover them as well without requiring a bypass.
