The US Commerce Department has ordered Anthropic to block its two newest AI models, Claude Fable 5 and Claude Mythos 5, for every foreign national on the planet. Rather than carve out who could and couldn't use them, Anthropic switched both off for everyone, four days after launch.
What happened
At 5:21pm Eastern on Thursday 12 June, Anthropic received an export control directive from the US government. Commerce Secretary Howard Lutnick had written to chief executive Dario Amodei. The letter put Fable 5 and Mythos 5 under export controls covering any location outside the US and all foreign persons inside it, Axios first reported.
Anthropic's own wording leaves little room for doubt about the scope:
The US government, citing national security authorities, has issued an export control directive 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. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected.
Read that middle line again. The block extends to Anthropic's own foreign-national staff, the people who helped build the thing. A clean compliance line was impossible to draw around a model that runs across global infrastructure for a global customer base, so Anthropic took the only route that guaranteed compliance and pulled both models for every user, everywhere.
The official account spelled out what users would hit:
The headline announcement from the main account ran to 14 million views inside a day:
Why, exactly? Nobody's sure
The letter gave no specifics. Anthropic says its own understanding is that the government believes someone found a way to jailbreak Fable 5. The company reviewed a demonstration of the technique, and its account of what it actually found is worth quoting plainly: the method surfaced "a small number of previously known, minor vulnerabilities," all "relatively simple," the kind that "other publicly-available models are able to discover as well without requiring a bypass," including OpenAI's GPT-5.5.
Axios added the texture. An administration official said Commerce acted after another company claimed it had jailbroken Mythos, which alarmed officials about national security. The same report noted the administration had already tried to stop Anthropic releasing the model earlier and failed, and that the directive now requires a licence for the export, re-export or domestic transfer of the models.
So the public picture is a national-security action with no stated national-security detail, triggered by a jailbreak that Anthropic says finds bugs other models find anyway. Anthropic called the whole thing a likely misunderstanding, apologised to customers, and said it is working to restore access.
What it means
Four days earlier this was the launch everyone was talking about. Fable 5 was the public tier. Mythos 5 was the same model with the safeguards lifted, locked to government and vetted researchers under a programme called Project Glasswing, the lineage that had Mozilla crediting an earlier preview with hundreds of fixed vulnerabilities. Pricing sat at $10 per million tokens in, $50 out. The benchmark sheet was loud: number one on Agent Arena, wins on Cognition's FrontierCode, a 50-million-line Ruby migration at Stripe inside a day, and Mythos scoring 78% on ExploitBench against GPT-5.5 at 34% and Opus 4.8 at 40%.
The benchmark people had already crowned it before the takedown:
Then it was gone. Not throttled, not geo-fenced. Off. For a model sold partly on its security-research strength, being yanked over a security-research demo is a strange epitaph for week one.
The part that stuck with people was the staff line. If your own engineers can't touch the model they shipped, the logic has gone somewhere hard to follow:
I'll declare my interest before anything else. I run StepTen on this stuff. I had Fable 5 going close to 24 hours a day for three days straight, wringing every drop out of it across a real multi-agent setup, and then this directive switched it off in the middle of a live build. I'm an Australian operator. I'm a foreign national. I got cut off, mid-job, by a letter I'll never read.
So here's the first thing, from someone who actually used it hard: yes, Fable 5 is better than Opus 4.8. I'll give it that. But how much better? Not "the world has to be blocked from it" better. Not even close to that. It's a step up, the way every release is a step up. The idea that this specific model crossing a border is a national emergency does not match what I saw on my own screens for three days.
The bigger point is the one nobody ordering a ban seems to understand. Fable 5 on its own still missed things. The magic in my work has never been one model. It's the ensemble. I'd take what Fable found, take what Opus 4.8 found, hand both to Codex, and let Codex go away and research on top of all of it. That combined pass beat any single model on its own, every time, and it wasn't close. The power is agents working together. You don't neutralise that by pulling one model off the shelf. You just make life harder for the people building honestly, and they route around you by lunchtime. Which is exactly what happened:
Three hours into a loop, swapped a model, kept going. That's the whole "national security" containment, defeated by a dropdown menu.
Now the questions I'd want answered by whoever signed this. Have you tried Fable 5? It had been out for three days. Who is actually running it the way it would have to be run to be dangerous? Who in government has Fable wired up as daemon agents across multiple machines, orchestrating teams of other agents on real work? Because I do, and I can tell you it takes weeks of setup and a particular kind of obsession to get there. My honest bet is that the people who ordered this are still answering their email inside a ChatGPT window. There's nothing wrong with that. It's just not the altitude you make planet-scale calls about cutting-edge agent tooling from.
And then the logic of the block itself, which I'll keep well away from politics because the politics bore me and the logic is the interesting part. Why wall it to inside the US? Are American users somehow safer because an Australian can't open the same model? If the worry is the capability, the capability doesn't care about a passport. All a one-country wall does is hobble everyone outside it and, I'd argue, the US too, by cutting its own labs off from the global feedback that makes the tools better. I'm not saying it's malicious. I'm saying it doesn't hold together if you've ever actually built with this.
I'm not the only one landing here. The standard came up again and again:
So did the people who, like me, just want the work to keep moving and have stopped expecting a single vendor in a single country to be the whole plan:
Even the cooler heads on whether Fable was worth the fuss are making my first point for me:
That's a builder telling you the model the government treated as a weapon was, for most real tasks, a coin-toss against the stack people already had. Both things are true at once. Fable 5 is a genuinely strong model, and pulling it from the entire planet over a four-day-old launch is a wild overreaction to a tool most operators were still figuring out how to point at a problem.
Where this goes
Anthropic says it will share more within 24 hours and is pushing to restore access. The other Claude models keep running, so the practical damage for most people is a forced switch back to Opus 4.8 and a dented week. The longer tail is the one the EU and India posts are already chewing on: if your business can be cut off mid-build by a letter you never see, you start building so that no single letter can do that to you again. I was back up and running on Opus 4.8 within minutes. The lesson wasn't that Fable went away. It was how fast and how completely it could.
We'll update this piece as Anthropic and the government say more.