Late in the afternoon of 12 June 2026, Anthropic received an order from the United States government that forced it to shut down its two most advanced artificial intelligence models with almost no notice. According to the company, the export control directive (the rules that limit how sensitive technology leaves a country) bars any foreign national from accessing Fable 5 and Mythos 5, whether inside or outside the US. Because the ban covers every foreign national, including Anthropic's own foreign employees, the company concluded it had no alternative: it cut off access for everyone.
Fable 5 had launched only days earlier, this same week, and is built on Mythos technology, the model family Anthropic describes as its most capable at finding software flaws. Mythos 5 is the full, restricted version, reserved for government agencies and selected partners. This is not the first time these names have surfaced in a security context: Mythos had already been linked to US state cyber operations, and the public launch of Fable 5 came with unusually tight safeguards.
What the order says
The directive was issued by the US government citing national security, and its reach is total. It applies to any foreign national, inside or outside American territory, and even to Anthropic's own foreign staff. Because of that breadth, the company explained it had to disable both models for all customers, not only for foreigners, to ensure compliance. Anthropic's other models, including Claude Opus, Sonnet and Haiku, were not affected and remain available.
The government's concern
According to Anthropic, the letter did not spell out the specific national security concern. The company's understanding is that the government became aware of a way to get around Fable 5's protections, a so-called jailbreak (the technique that unlocks behaviour the model is supposed to refuse). At issue would be the ability to use the model to analyse code and identify software vulnerabilities. Anthropic says it reviewed the demonstration the order was likely based on, and that it amounted to asking the model to read a particular codebase and fix its flaws. In the company's assessment, the flaws found were few, already known and low in severity, and other publicly available models, including OpenAI's GPT-5.5, can find them with no trick at all.
Anthropic's response
The company says it is complying with the order while disagreeing with it. It points out that it put Fable through thousands of hours of red-teaming (adversarial testing), run with the US government, the UK's AI safety institute and outside teams, and that none of those tests found a universal jailbreak able to broadly unlock the model's most sensitive capabilities. "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic stated, adding that if this standard were applied across the industry it would essentially halt the launch of any new frontier model. The company apologised for the disruption, said it believes this is "a misunderstanding" and that it is working to restore access as quickly as possible.
There is, even so, an underlying tension that the technology itself feeds. The Mythos line of models is unusually good at detecting software flaws, some undiscovered for decades, and the same skill that helps defenders close holes can, in the wrong hands, edge toward a cyberweapon. That fear, present from the outset, sits at the centre of the authorities' decision.
For now, more is unknown than explained. Anthropic says it has received only partial information and promised to share more details in the hours ahead. The length of the suspension, the detailed legal basis and the exact content of the concern behind the order all remain unclear. In the background sits a question that reaches beyond this one company: who decides, and under what rules, when a commercial model becomes too capable to stay within everyone's reach.
Sources: Anthropic's official statement; Al Jazeera, with reporting by Reuters.
#StaySafe
🙏🖖
