On June 12, 2026, two of the most capable AI systems available to businesses anywhere in the world went dark. Anthropic's Fable 5 and Mythos 5 models were pulled from all users, worldwide, after the US Commerce Department issued an export control directive giving the company 90 minutes to comply.
Here is what actually happened. Amazon researchers, testing Fable 5 shortly after its public launch, found that they could ask the model to read a piece of software code and identify flaws in it. That sounds benign. But Fable 5 is built on Mythos, a model Anthropic had specifically kept off the market because of its advanced ability to find and exploit software vulnerabilities. The worry was that this technique could be used to find weaknesses in systems and then attack them.
Amazon CEO Andy Jassy happened to be on a pre-scheduled call with White House officials about something entirely unrelated. He mentioned the findings. Within hours, Treasury Secretary Scott Bessent was briefed. The following day, the Commerce Department sent a letter to Anthropic CEO Dario Amodei: fix the flaw or remove the models. Amodei refused to pull them on those terms. By Friday evening, the order was issued and both models were offline.
Anthropíc's core argument is straightforward: the same technique works on OpenAI's GPT-5.5 and several other publicly available models, none of which have been banned. The company says the jailbreak is narrow, not universal, and that the vulnerabilities it uncovered were already publicly known and minor. Security researcher Katie Moussouris, who reviewed the original research, added that the flaw cannot actually be fixed without breaking the model's legitimate code-review function. More than 76 cybersecurity experts signed an open letter calling the ban counterproductive.
But Anthropic is also a company that spent considerable effort publicly describing its own models as potentially dangerous. When Mythos first launched, Anthropic argued the model was too powerful to release broadly and withheld it from the public. When Fable 5 shipped as a consumer-safe version, Anthropic highlighted the safeguards designed to contain Mythos's capabilities. That narrative, useful for signaling responsibility to regulators, turned into a liability. As one cybersecurity researcher put it: "If you describe your product as a munition in every press release, eventually a government takes you at your word."
This episode also did not come out of nowhere. Earlier in 2026, the Pentagon designated Anthropic a "supply chain risk" after the company refused contract terms that would have allowed its models to be used for autonomous weapons and mass domestic surveillance. That dispute is still in court. The Fable ban is the second major clash in a matter of months.
For businesses outside the US, the fallout is a sharper version of a risk that has been present for some time: dependence on a small number of American AI providers whose availability can be interrupted by a single government decision, with no advance notice. Allied governments including the UK, Canada, and Australia formally protested. A UK exemption request was turned down by the White House on June 17. French President Macron warned publicly that no country would purchase American AI if it could be switched off at any moment.
As of June 20, both models remain offline. Trump said at the G7 summit that negotiations with Anthropic are "going fine," but no restoration date has been set. Anthropic recently filed a confidential IPO prospectus, meaning this disruption lands at the worst possible time commercially.
For any organization using AI models provided by US companies, the lesson is plain: access to these tools is not guaranteed. The better operators are already thinking about what a three-week outage of their primary AI platform would cost them, and whether they have any backup.