Anthropic Remains at Odds With White House Over Fable 5 Ban

The Trump administration's June 12 export-control directive did not just restrict foreign access to Anthropic's most powerful AI models -- it took them offline for everyone. To comply with a directive from Commerce Secretary Howard Lutnick, Anthropic said it had to 'abruptly disable' access to both Fable 5 and Mythos 5 for all customers, according to Wired and corroborating reporting from Axios and NBC News. Roughly eleven days after the initial order, the models remained offline as Anthropic engaged in working-group meetings at the Commerce Department.

The trigger was a jailbreak reported to the government by a competitor described by multiple sources as Amazon. Anthropic's public position, reiterated in those Commerce Department sessions, is that the administration's response is disproportionate. The company said it had received only verbal notice of what it characterized as 'a potential narrow, non-universal jailbreak' and disagreed that a recall was warranted. Anthropic also noted that comparable capabilities are reportedly available in competing models, including OpenAI's GPT-5.5 -- a pointed implicit argument about selective enforcement.

The honest caveat is that the specific technical details of the jailbreak have not been made public, so neither Anthropic's characterization nor the administration's decision can be independently evaluated. What the reporting does not give you is clarity on what the Commerce Department would actually need to see to allow the models to return, or whether the working-group process operates on any agreed timeline.

The broader implication may matter more than this particular standoff's resolution. If a competitor's report to regulators can take a frontier model offline on short notice -- while the developer disputes the severity of the underlying issue -- that is a new kind of availability risk for anyone building on frontier AI APIs to price into their architecture decisions.