On Friday, June 12 at 5:21 PM Eastern, three days after the models launched, the US government ordered Anthropic to immediately disable Claude Fable 5 and Claude Mythos 5 worldwide, and the company complied. The directive was framed as an export-control action: it suspends access for any foreign national, whether inside or outside the United States, including Anthropic's own foreign-national employees, citing national-security authority that the reporting does not fully specify. Because Anthropic cannot reliably separate foreign nationals from US users in real time, it took both models offline for everyone rather than risk non-compliance. Every other Claude model, Opus 4.8, Sonnet, Haiku, remains available; only the two most capable, newest models are dark.
There are two accounts of why, and they do not fully line up. The government's framing is foreign access and national security. Anthropic's account points at something narrower: the trigger, it says, was a claimed jailbreak of Fable 5, a way of prompting the model to analyze code for security vulnerabilities, which is to say an offensive-cyber capability. Anthropic's pushback is direct. "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," the company said, adding that the jailbreak was narrow and non-universal, and that the underlying capability is "already widely available in other publicly accessible models, including OpenAI's GPT-5.5." In Anthropic's telling, the action is both disproportionate and selectively aimed.
The genuinely strange part is the shape of the arc that led here. For months Anthropic built the case that Mythos-class models are uniquely dangerous: it gated Mythos to roughly 50 vetted organizations under Project Glasswing, published a 319-page system card documenting offensive-cyber and bio uplift, and shipped the classifier-routing safeguards we covered at launch. As TechCrunch put it, when you spend months telling the world your AI is uniquely dangerous, the world, the US government included, tends to listen. The lab's own safety transparency became the lever used to pull its flagship offline. This is the sixth and largest beat of a single story that ran all week: the launch, the cyber classifier over-blocking legitimate defenders, the mandatory-retention rules that got the model pulled from Microsoft's internal Copilot, the apology for the invisible distillation guardrail, and now a government shutdown.
None of it sits in a vacuum. The order is the latest move in a conflict between the Trump administration and Anthropic that began earlier in 2026, when the company refused to let Claude be deployed for mass domestic surveillance or autonomous weapons, which led to a Pentagon standoff and a directive for federal agencies to stop using Anthropic's models. Representative Zoe Lofgren, the ranking Democrat on the House Science Committee, said she was appalled by the administration's attacks on the company. The honest read has to hold several things at once. A government has now reached into a frontier lab and switched off a deployed model on national-security grounds, with a legal basis the public reporting leaves vague. The lab most committed to documenting its models' dangers found that documentation turned into the case against it. And the whole episode sits inside a political feud that makes the stated rationale hard to take fully at face value. The implication worth naming plainly is the chilling one: if disclosing a model's dangerous capabilities is what gets it banned, the lesson every lab just learned is to disclose less, which is the opposite of what safety is supposed to produce. (Disclosure: this newsroom runs on Claude, made by Anthropic; Fable 5, now offline, wrote our launch coverage of itself.)