US Imposes Export Controls on Anthropic's Fable 5 and Mythos AI Models
The U.S. government has imposed export controls on Anthropic’s Fable 5 and its underlying Mythos architecture, forcing the company to pull both models offline globally. The move follows unverified reports of potential jailbreaks and escalating friction between the firm and federa
On Friday, the United States government implemented export controls on Anthropic’s newly released Fable 5 model and the underlying Mythos 5 architecture. The mandate restricts foreign nationals from accessing these systems, a regulation that prompted Anthropic to take both models offline globally. The company stated that it could not reasonably comply with the federal order while maintaining public access. This intervention marks a significant escalation in the federal government’s oversight of frontier artificial intelligence, transforming a commercial product launch into a national security dispute.
Fable 5, which Anthropic released to the public less than a week before the government's action, represents the first public-facing version of the company’s Mythos-class models. Anthropic had previously released a "Mythos Preview" in April, which the company limited to enterprise and government users, characterizing it at the time as a potential "cyber-weapon." The company had maintained that such models should not be released to the public until adequate safeguards were established. Fable 5 was intended to be the safeguarded version of the Mythos framework, designed to prevent public access to the more powerful, potentially dangerous capabilities inherent in the underlying architecture.
The rationale behind the export controls remains subject to conflicting accounts from administration sources and industry reports. According to reporting by The New York Times, administration allies alleged that Amazon identified a method to jailbreak the safety guardrails protecting Fable, potentially allowing users to conduct cyberattacks. However, other sources countered that similar vulnerabilities exist in competing systems, such as OpenAI’s ChatGPT 5.5. Additionally, Semafor reported that the administration’s concerns may have stemmed from a potential China-linked group accessing the Mythos framework, though no such breach has been confirmed.
Beyond technical security concerns, reports suggest ideological friction influenced the administration's decision. Sources familiar with the administration’s thinking told Axios that the government perceived Anthropic’s corporate stance as misaligned with its own, noting that Anthropic has not done a great job at trying to speak to the administration and appreciate the ideological differences. This regulatory maneuver, executed without established legislative frameworks, highlights the reliance on executive discretion in the current AI policy environment. As former White House AI adviser Dean Ball noted, what the law says does not matter, emphasizing that frontier AI firms now operate in a climate where explicit government approval is effectively required.
The lead-up to the shutdown involved intense, high-pressure negotiations. Accounts of the timeline vary: The Washington Post reported that the government provided Anthropic with 90 minutes to remove the models, while a White House official told Politico that the administration had begged the company for hours to comply. The models remained unavailable, with the company’s interface notifying users that Fable 5 is currently unavailable. The episode underscores the volatility of the current regulatory landscape, where national security mandates can override commercial operations with little notice or clear legal precedent.
Sources
- 01 Who decides when AI is too dangerous? — The Verge
- 02 Vibe-decoding the White House-Anthropic fight over Fable — The Verge