Why Fable 5 suddenly disappeared — the US government story, explained simply

One day Fable 5 was the exciting new thing in AI. A few days later it was gone — no warning, just a notice that it was no longer available. If you tried to use it and found nothing there, or if you saw confusing headlines about 'export controls' and 'jailbreaks', you're not alone. Here's what actually happened, explained as clearly as possible.

Fable 5 launched on 9 June 2026. It was Anthropic's most capable public model — a version of the more powerful, restricted Mythos 5 that was made available to everyone. Three days later, on 12 June, Anthropic received what's called an export-control directive from the US government. Export controls are rules that the US uses for sensitive technology — things it doesn't want leaving the country or reaching certain people. Normally these rules apply to things like advanced microchips or software before they're distributed. This one was different: it applied to a model that was already live and being used.

The directive said no foreign national — not citizens of specific countries, not 'people in adversary nations', but quite literally anyone who isn't a US citizen — could have access to Fable 5 or Mythos 5. That's an enormous group of people. Think about how many non-US citizens use AI tools: researchers, students, professionals, people in every country. And here's the thing about Anthropic's own team: many of their employees aren't US citizens either. There is genuinely no practical way to let 'only US citizens' use a global product that employees across the world also rely on. So Anthropic made the decision to switch both models off entirely for all users — including US ones — while it works to sort this out.

The official reason given was national security. Specifically, the concern was about a 'jailbreak' — a way of getting around a model's safety rules. Here's the honest part, and it's worth being straight with you about it: the full reasoning hasn't been made public. Anthropic says it was only told verbally about a 'potential narrow, non-universal jailbreak' — meaning it wasn't given a formal written explanation. The company reviewed what it was shown and genuinely disagrees that the issue is serious enough to justify pulling the model. Anthropic even pointed out that another major AI company's model, OpenAI's GPT-5.5, apparently has the same underlying capability — but that model is still running normally. So right now, there are real questions about the reasoning that neither the government nor anyone else has answered publicly.

For most people using Anthropic's products day-to-day, no — not at all. Every other Claude model is completely fine: Claude 3.5 Sonnet, Claude Opus 4.8, and the standard Claude apps all work exactly as before. The only things that are switched off are Fable 5 and the restricted Mythos 5, which were only recently launched and which most casual users hadn't yet made a core part of their workflow. If you never used those two specific models, your week looks exactly the same.

Where it matters more is for developers and researchers who had started building on Fable 5 or Mythos 5 through the API, and for users who specifically sought out the more capable model after launch. For them, it's a genuine disruption — the most capable model they'd started using has simply vanished mid-build. Anthropic has said it's working to restore access and is committed to doing so, but there's no firm date and the situation is still developing.

Even if you're not personally affected by Fable 5 disappearing, this episode matters for a reason worth knowing about. Before now, the US government's export-control rules for AI were mostly about preventing advanced AI chips from being shipped to certain countries, or limiting how model weights (the underlying files that make a model work) could be distributed. This is the first time those same legal tools have been used to reach through a deployed product — a model already live and in use — and switch it off while people were actively using it. That's a different kind of intervention, and it sets a precedent for how governments might regulate AI going forward.