Anthropic and White House remain at odds over Claude Fable 5 export controls
The U.S. administration kept export restrictions on Anthropic's Claude Fable 5 after talks in Washington, while the company maintains the model's safeguards are adequate.
U.S. officials concluded a series of meetings with Anthropic on Monday without lifting the export controls placed on the company's Claude Fable 5 model last week. The administration says the model’s guardrails can be bypassed, potentially exposing the more powerful Mythos capabilities, while Anthropic argues the risk has been overstated.
Anthropic representatives, including co-founder and chief compute officer Tom Brown and head of external affairs Sarah Heck, met with Commerce Department officials, researchers from the Center for AI Standards and Innovation, and the Office of the National Cyber Director. Commerce Secretary Howard Lutnick participated by conference call from the G7 summit in France.
"Both parties are working quickly to get this resolved," an Anthropic spokesperson said in a statement to WIRED.
The White House declined to comment. The Commerce Department indicated it may allow Fable 5 to resume consumer access if Anthropic can fully address the alleged jailbreak vulnerabilities, though no timeline was provided.
The dispute follows earlier concerns raised by Amazon, a major investor in Anthropic, after its CEO contacted Treasury Secretary Scott Bessent about possible model weaknesses. The National Security Agency subsequently assessed that the guardrails could be stripped, prompting the export restrictions.
Security researchers have contested the administration’s assessment. An open letter signed by several experts argued that the controls remove valuable tools from defenders without demonstrable risk, noting that similar capabilities exist in other foundation and open-source models. Katie Moussouris, founder and CEO of Luta Security, said the reported issue did not constitute a full jailbreak and that guardrails function more as speed bumps than absolute barriers.
Investors are monitoring the situation, with some suggesting the government may be singling out Anthropic. Industry observers note that the episode could set a precedent for how AI labs engage with U.S. regulators on advanced model releases.
"The events over the weekend … are informative for everyone that the [U.S.] government would be willing to take these steps," said Aidan Gomez, CEO of Cohere.
