Breaking: US Government Orders Anthropic to Disable Mythos and Fable 5
Just now—literally just now—the US government ordered Anthropic to shut off access to Mythos and Fable 5. At 5:21 this morning Anthropic published a statement, which I've translated.
The gist: the government ordered Anthropic to block any foreign national from accessing these two models, on the grounds that they are simply too capable. For now, Anthropic's only way to comply is a blanket shutoff—access to Fable 5 and Mythos is suspended for every customer.
Anthropic itself says it doesn't understand the order. And this is the first head-on conflict between a frontier AI vendor and the government.
Just now—literally just now—the US government ordered Anthropic to shut off access to Mythos and Fable 5. At 5:21 this morning Anthropic also published a statement about it, which I’ve translated into Chinese.
What happened
The gist of the statement: the government ordered Anthropic to block any foreign national from accessing these two models, on the grounds that they are simply too capable. For now, Anthropic’s only way to comply is a blanket shutoff—access to Fable 5 and Mythos is suspended for every customer.
The government’s reasoning, and Anthropic’s response
Anthropic itself says, plainly, that it doesn’t understand the order.
The government’s reasoning is that people have already found ways around Anthropic’s safety guardrails to run cyberattacks, and even research into biological weapons.
Anthropic’s view is that the evidence the government provided doesn’t hold up. As it stands, that reasoning is not grounds for recalling a commercial model deployed to hundreds of millions of people. Anthropic acknowledges the government does have the authority to block unsafe deployments, but says any such move has to go through a transparent, fair, technically grounded statutory process—and this demand doesn’t meet that bar.
Anthropic also said this may be a misunderstanding, and that access to the models might be restored soon through dialogue.
A gut check on how strict Fable 5’s guardrails are
I’ve been using Fable 5 myself, and I find its guardrails extremely strict. How strict?
ElevenLabs is the small model I normally use to generate music. When I called ElevenLabs through Claude Code, purely because the name contains “Labs,” it refused to let me use Fable 5 to write or deploy the related code.
Guardrails that strict, and the model still got pulled offline today over this.
My read
As I see it, this is the first head-on conflict between a frontier AI vendor and the government, and to some degree it signals government export controls on top-tier models.
Separately, Anthropic has now refreshed this week’s quota. I don’t know whether that’s related.
What’s your take on all this? Let’s talk in the comments.
Source: EP0039_audio.mp3 · ASR model gemini-2.5-pro (chunked parallel) · full text of the original recording
[00:00] Just now, literally just now, the US government ordered Anthropic to shut off access to Mythos and Fable 5. At 5:21 this morning Anthropic also put out a statement about it, and I translated it into Chinese so you can all read it. The gist is that Anthropic is required to block any foreign national from accessing these two models, because these two models really are too capable. So to stay compliant, all Anthropic can do right now is a blanket ban—it has suspended access to Fable 5 and Mythos for every single customer.
[00:25] Anthropic’s own take is that it doesn’t get this at all. The government’s reasoning is that someone can already get around Anthropic’s safety guardrails to run cyberattacks, and even some bioweapons research. Anthropic itself thinks the evidence the government provided doesn’t hold up, and that the current rationale shouldn’t be grounds for recalling a commercial model deployed to hundreds of millions of people. Anthropic does think the government has the authority to block an unsafe deployment—but it has to go through a transparent, fair,
[00:51] technically grounded legal process, and this particular order doesn’t meet those principles. Anthropic also said this may be a misunderstanding, and that talking it through could get access to the model back soon. From the way I’ve been using Fable 5, I’d say its guardrails are extremely strict. How strict? ElevenLabs is a small model normally used to generate music, and when I called ElevenLabs through Claude Code—just because that word has “Labs” in it,
[01:17] it would block me from using the Fable 5 model to write and deploy any of the related code. Guardrails that strict, and yet today it went offline over this. Which makes this, I think, the first head-on clash between a frontier AI vendor and the government. And to some degree it also signals the government moving to control exports of top-tier models.
[01:42] Anthropic has now refreshed this week’s quota. I don’t know whether that has anything to do with this. What do you all think about it? Let’s talk in the comments.