Washington Pulls Claude Fable 5 and Mythos Offline

I build with Claude Code every day, so my honest first reaction to this was a cold one: a model I lean on for real work can be switched off by government order, overnight, with zero input from me. On Friday, June 12, 2026, at 5:21 PM ET, Anthropic complied with a US government directive to disable Claude Fable 5 and Claude Mythos 5 worldwide. The order cited national security and export-control concerns. The company's other models stay online — but its two most powerful ones went dark.

Here's a quick map of what got pulled. Mythos 5 is Anthropic's most capable model — previewed back in April and never opened to the public. It ran behind a controlled program called Project Glasswing, available to roughly 50 vetted organizations with names like Amazon, Apple, Google, Microsoft and CrowdStrike, precisely because it was unusually good at finding software vulnerabilities. Fable 5 was the public-facing version: same underlying brain, plus guardrails meant to block high-risk areas like cybersecurity and biology. It had been available for exactly three days before the plug came out.

The trigger is thinner than the consequence

The stated reason was a "potential narrow, non-universal jailbreak" of Fable 5 — in plain language, someone got it to read code and point out the flaws in it. Anthropic is not hiding how it feels. The company wrote that it "disagree[s] that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people." Its argument is that the same capability lives in rival models like OpenAI's GPT-5.5, and that security professionals use exactly this kind of code analysis defensively, every single day.

There's a bitter irony underneath. Sam Altman had earlier needled Anthropic for what he called "fear-based" marketing around how dangerous Mythos supposedly was. Whether or not that's fair, the same steady drumbeat about the model's power is part of what drew the government's attention in the first place. You can spend a year telling the world your AI is uniquely dangerous and then act surprised when someone with subpoena power takes you at your word.

Why should you care if you've never touched Mythos and never will? Because this is the first time I can remember a frontier model being recalled like a faulty car part, by a government, mid-deployment. The lesson for anyone who has wired an AI tool into daily life — work, study, a side project — is that the assistant sits on top of a supply chain you don't own. A model can get more restrictive, change its behavior, or vanish entirely between a Tuesday and a Friday, and your routine goes with it. We've started treating these things like the power grid: always on. They are not. They are commercial products living inside politics.

My guess is this won't be the last recall of its kind. The more capable these models get, the more they look like infrastructure that regulators feel entitled to switch off. That's not a reason to stop using them — I'm not going to — but it is a reason to stop treating any one of them as permanent.