Last week I wrote about Claude Fable 5 twice here: first about the launch, then about an afternoon of real work with it. I signed off with “I cannot wait to use it more.” That wait just got longer than I expected. Since June 12, Fable 5 can no longer be used.
What happened
At 5:21pm ET that day, Anthropic received an export-control directive from the US government, citing national security. The order: suspend all access to Fable 5 and Mythos 5 for any foreign national, inside or outside the US, including Anthropic’s own foreign-national employees. To comply, Anthropic had to disable both models for everyone. All other Claude models keep working.
The trigger, as far as is known: the government believes it has seen a method to “jailbreak” Fable 5, to bypass its safety guardrails. Anthropic says it reviewed that demonstration and recognized a handful of already-known, minor vulnerabilities in it. Simple ones, they write, that other publicly available models surface too, without needing any bypass at all.
Two stories, one side
Here I have to be honest about what we do and do not know. So far we mostly have Anthropic’s version. The company is complying with the order while openly disagreeing with it. The heart of their objection: one narrow, possible jailbreak is, in their view, no reason to recall a commercial model used by hundreds of millions of people. Apply that standard broadly, they argue, and you could not ship a single frontier model again.
Anthropic says it spent thousands of hours testing before launch, alongside the US government itself, the UK AISI and outside parties, and that nobody found a universal jailbreak. Their position was always that perfect resistance does not exist, that every model in the industry is vulnerable to narrow bypasses, and that the answer is defense in depth plus monitoring. The fact that they wrote this down in advance is their strongest card in this conversation.
What is missing is the government’s reasoning. The order names no specific concern. That makes it hard to judge who is right here. It could be a misunderstanding, as Anthropic claims. It could also be that the government sees something a blog post does not capture. So I will stick to what is settled: the model is offline, the company is working on restoring access, and more information was promised within a day.
Why this matters to the builder
If you only open a chat window now and then, this is a news item. If you build work on these models, as I do, it is a test of an assumption.
In my test run I asked Fable 5 itself whether another model could have done the same job. The answer was: largely, yes. The scripts, the gates, the rules in my workflow were deliberately built model-agnostic. Back then that felt like a footnote. Now it turns out to be a safety net. A model can vanish overnight, for reasons that sit outside you and outside the company. Lock your entire stack onto that one model and you grind to a halt. Build your system around it and you fall back to Opus 4.8 and keep going.
That is exactly the lesson I wrote down last week, now tested sooner than I expected: the system does more than the model. The magic was never in that one model number. It is in the rules, the memory and the chain around it. A recall hurts, but it does not end the work.
And honestly: I am sorry to see it go. But I also understand it. When a government doubts whether a capability is safe, a pause is the sensible answer, even if it costs me my favorite tool of the moment. I hope it comes back, and I think it should. Not because the discomfort needs to disappear, but because the only way we learn to handle models at this level is by actually using them and finding their limits in the open. Hiding them away teaches no one anything.
Whether Fable 5 returns, and when, I do not know. I am following it, and the moment there is more, you will read it here.
This piece also appeared in Dutch: Fable 5 uit de lucht: de overheid trok het sterkste model terug.