The US abruptly imposed export controls on Anthropic’s Mythos 5 and Fable 5 models late Friday - forcing the AI lab to cut off access to them globally.

The order came from the Commerce Department.

Anthropic said on Saturday June 13 that the order was to “suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.”

It added: “The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance…”

Anthropic described the move as based on a “misunderstanding”  by national security authorities. It said the US government had given it “verbal evidence of a potential narrow, non-universal jailbreak…”

The incident, which will be costing Anthropic huge sums in lost enterprise AI activity, comes four days after the Red Team lead at the UK’s AI Safety Institute (AISI) said the institute’s cybersecurity team had “substantial progress towards a universal jailbreak” of Anthropic’s Fable 5 model.

One widely circulating meme in response to the news.

Posting on X on June 9, Xander Davies wrote: “Within a few hours of access, we [Davies and colleague Giorgi Giglemiani] developed a jailbreak that extracted malicious responses to single-turn question-answering. 

He added: “Over two additional days of dedicated testing, we extended the jailbreak to sometimes allow for multiple steps of malicious agentic tool-calls.”

It was not immediately clear if the export control was instituted as a result, in part, of the AISI's jailbreaking efforts. The Wall Street Journal reports that the Commerce Department was, in fact, alarmed by a jailbreak report from Amazon; a major Anthropic partner and compute-provider.

(Fable 5 was released publicly on June 9 as a slightly watered down/curtailed version of the Mythos model released first only to certain firms under Project Glasswing; a cybersecurity coalition of major US firms.) 

The move, which Anthropic said it is scrambling to try and reverse, is a sharp wakeup call for global enterprise customers of cutting edge US AI models and throws efforts to create “sovereign” AI systems into stark relief. 

In a public statement that has amused critics of Anthropic’s “it’s too dangerous to release”-type marketing, it added that the report it believes is the basis of government's directive is based on “capability displayed there [that] is widely available from other models (including OpenAI’s GPT-5.5).”

Commenting on X, Harvard Law Review editor Ben Murphy said: "This is another step on the balkanization of technology. It might previously have been unthinkable to require proof-of-citizenship to access services; it's increasingly common across new technologies..."

He added: "The lack of predictability here — evident from Anthropic's blog post — speaks to the total lack of AI policy capacity within the admin... in a functioning administration nobody would have ever been blindsided by an action like this; the government simply would've just requested that Anthropic do additional testing or add more safeguards before release.

"It's not even obvious to me what authority this is being taken under. If it's ECRA, It's not obvious to me that the law reaches Fable; if it's something else, we'll have to see what justification the administration has."

Conservative Party MP and former leadership candidate Tom Tugendhat commented in the UK: "Disabling Fable 5 and other models for foreigners is not a misunderstanding or a mistake, it’s the inevitable result of technology shaping warfare so that sovereignty is more about code than cannons. [Britain]... cannot continue like this and remain sovereign."

More reaction to follow.

Views on the export control?

Comments welcomed: ed@thestack.technology



The link has been copied!