Fable 5 launched June 9, was switched off June 12 by a US export-control order, and returned July 1 with a new safety classifier. Mythos 5 did not come back the same way. Here is the full timeline, what changed on the way back, and why your Fable prompts need re-checking.

Fable 5 is available again. The US Department of Commerce lifted its export-control order on June 30, 2026, and Anthropic redeployed the model on July 1 to users worldwide. If you are reading this because you still see "Claude Fable 5 is currently unavailable," that message is stale on your end, so sign out and back in or check that your plan includes it.

Mythos 5 is a different story, and the difference is the part most coverage skipped. It did not return alongside Fable. More on that below.

I wrote the original version of this post while both models were dark and nobody had a return date. I have kept the history intact, because how this happened is still the most interesting part and it explains what changed on the way back. What follows is the timeline, why an order aimed at foreign nationals disabled the model for everyone, the jailbreak that set it off, what Anthropic changed before switching it back on, and what to do with the prompts you re-tuned during the outage.

What happened, in order

Fable 5 and Mythos 5 launched on June 9, 2026. They were the most capable models Anthropic had ever released to the public. Three days later they were gone.

According to Anthropic's statement, the directive arrived on June 12 at 5:21pm ET. The reporting attributes it to Commerce Secretary Howard Lutnick. By that evening both models were disabled globally. The order was an export-control measure: it required Anthropic to "suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States."

So this was a government action, not a product decision. That distinction matters, because it changes everything about how and when the model can come back.

Why an order about foreign nationals took the model down for everyone

The directive technically targeted foreign nationals. The shutdown hit every user. That gap confused a lot of people, so it is worth explaining the mechanism.

Anthropic does not verify the nationality of each person sending a request in real time. There is no live filter that can let a US citizen through while blocking a foreign national on the next request to the same model. Faced with an order it had to comply with immediately and no way to enforce it selectively, the company took the only option that guaranteed compliance. In its own words, "the net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance."

That is why a US citizen in Ohio and a developer in Berlin both see the same "currently unavailable" message. The block is all-or-nothing because the enforcement tool is all-or-nothing.

The jailbreak that triggered it

The government's stated concern was national security, and the specific trigger was a demonstrated way to get around Fable 5's safeguards. Anthropic reviewed a demonstration of what it called "a narrow, non-universal jailbreak." The method involved asking the model to read a specific codebase and fix any software flaws in it.

That sounds harmless, and in most contexts it is exactly the kind of work people want a frontier model to do. The security worry is the flip side of the same skill. A model good enough to find and fix flaws in a codebase is also a model good enough to find and exploit them, which moves it close to offensive cybersecurity capability. Reporting around the launch noted that Fable 5 had caught real bugs that GPT-5.5 and Opus 4.8 missed, so the capability was not theoretical. Anthropic was careful to frame the jailbreak as narrow rather than universal, meaning it affected specific tasks rather than unlocking the model wholesale.

Where Mythos 5 fits in

If you are new to the naming, Mythos 5 is the same underlying model as Fable 5 with the safeguards lifted, released to a small set of vetted partners. Fable 5 is the public version with safety classifiers watching the conversation. Because they share the same core capability, the directive named both, and both went dark together. Anyone who had partner access to Mythos 5 lost it at the same moment everyone else lost Fable 5.

How Fable 5 came back

The hold lasted eighteen days. Commerce lifted the export controls on June 30, Anthropic announced it was restoring access, and Fable 5 was live again on July 1 across the Claude Platform, claude.ai, Claude Code, and Cowork, for Pro, Max, Team, and select Enterprise plans. Access on Amazon Web Services, Google Cloud, and Microsoft Foundry followed.

Usage was metered on the way back rather than thrown wide open. For the first week, through July 7, eligible subscribers could spend up to half of their plan's weekly usage limit on Fable 5. After that it moved to usage credits, which is where it sits now.

The prediction in the original version of this post was half right. It was a regulatory hold that lifted when the regulator was satisfied, and Anthropic did close the jailbreak before relaunching. What I got wrong was the shape of the return. I expected a staged reopening starting with verified US enterprise accounts, and instead it came back globally on day one.

What changed before it was switched back on

Anthropic did not simply flip the same model on again. It trained a new safety classifier aimed at the bypass technique that caused the problem, and reports it catches those attempts in more than 99% of cases. The company also worked with the US government and industry partners on a shared jailbreak framework rather than treating the fix as its own patch.

That number is Anthropic's own, measured against attacks they knew to test for, so read it as a strong claim rather than a settled fact. The practical consequence is what matters to you: Fable 5 today refuses differently than Fable 5 did on June 9. A new classifier watching a conversation changes where the line sits, and prompts that cleared the old one can trip the new one. If you built anything against launch-week Fable behaviour, it is worth re-checking rather than assuming it still runs.

Mythos 5 did not come back the same way

This is the question I still get asked most, and the answer is more specific than most coverage suggested.

The Commerce order was lifted for both models, so headlines reasonably read "Fable and Mythos restored." Only one of them returned to the public. Anthropic enabled Mythos 5 for a set of US organizations on June 26, and access since then expands through Glasswing, its cybersecurity program that gives selected organizations frontier-model access for defensive security testing.

So Mythos 5 is not off, and it is also not something you can go and use. It is gated behind a vetted program with a specific purpose, and the company has said it continues working with the US government to widen access to more domestic and international partners. If you are not in an organization doing defensive security work that qualifies for Glasswing, the practical answer to "when do I get Mythos 5" is that this is not the path it is on. Fable 5 is the version available to you, and for ordinary work it is the one you want anyway, since the difference is safeguards rather than raw capability.

What to do with your prompts now

Most people moved their prompts twice. Once onto Opus 4.8 or Sonnet 4.6 when Fable went dark, and again back toward Fable after July 1. Each of those moves quietly changed what good prompting looks like, and the second move is the one people skipped, because coming back to a model you used before feels like it should need no work.

It does. A prompt reshaped for Opus 4.8 during the outage carries Opus-specific scaffolding that Fable 5 does not need, and Fable can read the extra structure as noise. Running it now costs you output quality and tokens at the same time, and it shows up as answers that are slightly flatter than you remember without any obvious cause.

The new safety classifier is the second reason to re-check. Anything you wrote against launch-week Fable behaviour was written against a different boundary. That is not about doing anything shady; ordinary prompts in security, medicine, and law sit near classifier lines and get caught by changes like this.

That re-check is what FixMyPrompt is built for. Paste the prompt and it scores against the model you are actually running now, flags the shaping that belonged to a different model, and rewrites it around what this one responds to. Three free checks a day, no signup.

The short version

Claude Fable 5 went dark on June 12, 2026 because a US export-control directive barred foreign-national access and Anthropic, unable to filter by nationality in real time, disabled it for everyone to stay compliant. The trigger was a narrow jailbreak that had the model read a codebase and fix its flaws, which raised offensive-cybersecurity concerns. Commerce lifted the order on June 30 and Fable 5 returned globally on July 1, with a new safety classifier Anthropic says blocks the bypass technique in over 99% of cases. Mythos 5 did not return publicly; it was enabled for a set of US organizations on June 26 and expands only through the Glasswing defensive-security program. If you moved prompts to Opus 4.8 during the outage and moved them back, re-check them, because the scaffolding you added for Opus works against you on Fable and the refusal boundary moved.

Claude Fable 5 Is Back: What Happened and What Changed