In brief
The Commerce Department ordered that Fable 5 and Mythos 5 be withheld from all non-US persons, including the company’s own foreign staff, forcing a global shutdown.
Launched on 9 June, the systems were disabled on 12 June, leaving users worldwide without access to Anthropic’s most advanced AI.
A reported bypass method allowed the model to analyse large codebases and flag vulnerabilities; officials feared exploitation by hostile actors against US critical infrastructure.
The company says the same results can be achieved with other models like GPT-5.5, that extensive red-teaming proved Fable 5’s safeguards, and that no actual harm has been demonstrated.
Anthropic says it received only a verbal report of a narrow jailbreak, with no further specifics from the administration, calling the action a misunderstanding.
The episode fuels debate about US export controls on frontier AI and the interplay between national security and competition among American firms.
- 9 JunAnthropic launches Fable 5 and Mythos 5 to the public.
- 12 JunUS Commerce Department issues an export directive blocking the models for all foreign nationals.
- 12 JunAnthropic shuts down global access to Fable 5 and Mythos 5 for all users.
The export control directive
On 12 June, the US Commerce Department issued an emergency export control directive that forbids Anthropic from making its latest models available to any foreign national, whether in the United States or abroad, including the company’s own non-US employees. Because Anthropic’s services are global and do not verify citizenship at login, the firm said it was forced to pull Fable 5 and Mythos 5 for every customer.
“The ultimate effect of this directive is that we are suddenly forced to turn off Fable 5 and Mythos 5 for all our customers.”
What triggered the ban
The models, launched on 9 June, had quickly drawn attention for their coding and autonomous-task capabilities. Fable 5 was positioned as a direct rival to OpenAI’s GPT-5.5 and Google’s Gemini, while the Mythos class had already raised concerns internally. An earlier preview version was deemed by the company itself as
“too powerful to be made publicly available”
The jailbreak that alarmed Washington
A central factor was a report that reached the Commerce Department detailing a method to bypass Fable 5’s safety classifiers. Once jailbroken, the model could analyse large codebases and pinpoint critical weaknesses. Officials feared hostile hackers or state-backed groups could use that capability against US critical infrastructure, including energy, finance, and telecommunications. A researcher operating under the pseudonym "Pliny the Liberator" had previously claimed to have bypassed Fable 5’s safeguards and posted the system prompt on GitHub.
Silicon Valley pushes back
Anthropic vigorously disputed the government’s reasoning. The company said the jailbreak technique was narrow and non-universal, amounting essentially to asking the model to review a codebase and fix bugs.
“We analysed the specific technique and found that the vulnerabilities it identified are relatively simple and can also be detected by other publicly available models without any bypass.”
What remains online
All other Anthropic models, including Opus 4.8, remain unaffected. The company stated it had received no details from the administration beyond a notification from Commerce Secretary Howard Lutnick that the two models would be placed under export controls. The episode has already triggered broader debate about unilateral export restrictions on frontier AI and the blurring line between safety enforcement and competitive dynamics among America’s top AI labs.