Claude Fable 5 Unavailable: The Ban and Your Plan

Why Claude Fable 5 went offline: a June 12 export order

Fable 5 from launch to return (June–July 2026)

Dates from Anthropic's June 12 statement, its redeployment post and the Claude Help Center.

Jun 9Fable 5 and Mythos 5 launch Fable 5 included on paid plans through June 22

↓

Jun 12Export control directive arrives Anthropic turns both models off for all users

↓

Jun 26Mythos 5 approved for US organizations Fable 5 still offline

↓

Jun 30Export controls lifted on both Fable 5 and Mythos 5

↓

Jul 1Fable 5 back worldwide with a new safety classifier

↓

Jul 19Plan promotion ends access now depends on the plan

Claude Fable 5 had been public for three days when it disappeared. The cause was a government order about who may use the model, and Anthropic's only way to obey it was to take the model away from everyone.

Who the export control directive covered

An export control limits who may receive a technology. The directive Anthropic received at 5:21pm ET on Friday, June 12, 2026, cited national security and set that limit by nationality, applied this time to a hosted AI model.

It covered every foreign national, whether they were in the United States or abroad, and it named Anthropic's own foreign national employees. Both Fable 5 and Mythos 5 were in scope. Anthropic said the letter itself gave no specific detail about the security concern.

Why Anthropic switched it off for every user

The order took effect the moment it arrived. Anthropic had no reliable way to check a user's nationality in real time, so it could not keep US nationals on Fable 5 while blocking everyone else.

Anthropic therefore disabled both models for all customers, including Americans. That is why the model vanished for users in the US as well as abroad, even though the order itself covered only foreign nationals.

Which Claude models kept working during the ban

Only Fable 5 and Mythos 5 went dark. Every other Anthropic model stayed available for the whole period.

For most people the practical effect was one missing option in the menu. Anyone who had built work around Fable 5 had to move it to one of those models until the ban ended.

The Fable 5 jailbreak finding and Anthropic's response

The finding behind the directive, as Anthropic described it

From Anthropic's June 12 statement and its June 30 redeployment post.

Who reported itAmazon researchers

What the prompt didAsked the model to read a codebase and fix its flaws

What it foundA small number of known, minor vulnerabilities

Other modelsOpus 4.8, GPT-5.5 and Kimi K2.7 found the same bugs

The fixA new classifier that blocks the technique in over 99% of cases

The directive rested on a jailbreak: a way of wording a request so that it gets past the safeguards a model normally applies. The government saw a security risk in it, while Anthropic argued that the same capability was already available from weaker models.

What the government said the jailbreak does

According to Anthropic, the government's evidence was verbal and described one narrow technique. It amounts to handing the model a specific codebase and asking it to find and fix the software flaws in it.

The same skill that finds a flaw can show how to exploit it, which is why safety systems treat defensive security work as a borderline case. Anthropic reviewed a demonstration. By its account, the demo surfaced only a few previously known, minor vulnerabilities that other public models can find without any bypass.

Anthropic's case against recalling the model

Anthropic complied with the order and said so in the same statement where it disagreed with it. Its objection was one of proportion: a narrow potential jailbreak had led to the recall of a commercial model used by hundreds of millions of people.

The company also pointed to its pre-launch testing. The US government, the UK AI Security Institute, private firms and internal teams had spent thousands of hours trying to break Fable 5's safeguards. None had found a universal jailbreak, meaning one that unlocks cyber capabilities broadly.

What the Amazon report turned out to show

The June 30 post filled in what the first statement could not. The report behind the directive came from Amazon researchers, and in one case the model had produced code showing how a vulnerability could be exploited.

Anthropic then ran the same test on other models. Opus 4.8, GPT-5.5 and Kimi K2.7 identified the same vulnerabilities, and every model it tried, down to Haiku 4.5, could write the same exploit demonstration.

When Fable 5 came back online: July 1, after the ban

Where Fable 5 and Mythos 5 returned

From Anthropic's redeployment post (June 30, updated July 1, 2026).

July 1Fable 5, all countries Claude.ai, Claude Code, Claude Cowork, Claude Platform

↓

After July 1Fable 5 on cloud platforms AWS, Google Cloud, Microsoft Foundry

↓

LimitedMythos 5 a set of US organizations, approved June 26

The ban ran from June 12 to June 30, and the return came in stages. Fable 5 itself was back for general users the next day, but it did not come back unchanged.

Export controls lifted on June 30

Anthropic announced on June 30, 2026 that the export controls on both models had been lifted. In the two weeks before that, it had reviewed the Amazon report with the government and built a fix.

For anyone who had been waiting, this was the end of the ban. From that point on, "Fable 5 is unavailable" no longer meant the model was switched off for everyone.

The July 1 return and the new safety classifier

Fable 5 returned on Wednesday, July 1, in every country, on Claude.ai, Claude Code, Claude Cowork and the Claude Platform. It came back with an automated safety check that runs on every request and can block it before or during the answer.

The classifier added for the return blocks the technique from the Amazon report in over 99% of cases. When it blocks a request, the request goes to Opus 4.8 instead and the user is told. Anthropic warned that the classifier also flags harmless requests more often during routine coding and debugging.

Mythos 5 and the cloud platforms

Mythos 5, the version with fewer safeguards for defensive security partners, came back first but to a smaller group. After a US government approval on June 26, Anthropic restored it for a set of US organizations.

Fable 5 on AWS, Google Cloud and Microsoft Foundry was set to follow "as quickly as possible" rather than on July 1. The full account of the return is in Is Claude Fable 5 Back? Yes, With New Safeguards.

Why Fable 5 is still unavailable: your Claude plan

How each Claude plan gets Fable 5

From the Claude Help Center article "Claude Fable 5 on your plan", checked August 31, 2026.

FreeNot available

Pro, Team standardUsage credits from the first message

Enterprise standardUsage credits, only if the organization turns them on

Max, premium seatsUp to 50% of weekly limits, then credits or another model

API, usage-based EnterpriseStandard API rates

With the ban gone, your plan is the first thing to check if you still cannot pick Fable 5. Two things decide it: whether the plan includes Fable 5 at all, and whether usage credits are switched on. Usage credits are money added on top of a subscription to pay for extra use.

Free and Pro plans after the July 19 change

The Free plan has no Fable 5 at any point. Pro included it only for limited periods, most recently under a promotion that ended on July 19, 2026 at 11:59:59 PM PT.

Since that date, Fable 5 on Pro and on standard Team seats runs only on usage credits. If credits are off, the model is effectively unavailable. Pro and standard Team seats may be eligible for a one-time credit meant to ease the change.

Max plans and premium Team or Enterprise seats

On Max and on premium seats in Team and seat-based Enterprise plans, Fable 5 is part of the subscription. It can use up to 50% of the weekly usage limit at no extra cost, and it uses that limit faster than other Claude models.

The 50% is a cap inside the weekly limit, not extra room on top of it. Once you hit it, you can keep going on usage credits or switch to another model until the week resets. Standard seats on Enterprise plans get Fable 5 only if the organization has enabled credits.

Claude Code and Cowork version requirements

Your plan is not the only requirement. Claude Code supports Fable 5 only from version 2.1.170 onward, so an older install cannot use it at all.

Claude Cowork needs the latest version of Claude Desktop. On the web, desktop and mobile apps, Fable 5 appears as "Fable 5" in the model picker.

When Claude switches from Fable 5 to Opus: flagged requests

What happens to a flagged Fable 5 request

From the Claude Help Center article on model switching with Fable 5, checked August 31, 2026.

1You send a message to Fable 5 classifiers check every request

↓

2The request is flagged offensive cyber may go to Opus 4.8, dual-use biology to Opus 5 (current targets)

↓

3Opus answers in the same chat a notice says the model switched

↓

4The picker stays on Opus until you choose Fable 5 again

Sometimes the reply to a Fable 5 message comes back labeled Opus. That is the safety classifier from the July 1 return at work, and it is a different symptom from the model being unavailable.

What the Fable 5 safety classifiers block

Every request to Fable 5 passes through classifiers first. Requests about offensive security techniques, such as building exploits or malware, may be handed to another model, currently Opus 4.8, and requests in dual-use biology, chemistry and life sciences currently go to Opus 5.

Fable 5 can still help with routine security tasks, but Anthropic tells users to expect frequent fallbacks in that area. The biology classifier was last updated on August 6, 2026, and attempts to pull out the model's summarized reasoning are blocked as well.

The setting that stops automatic model switching

Switching is on by default the first time you pick Fable 5. After a switch, the model picker stays on Opus for the rest of that conversation. Choosing Fable 5 again can trigger the same fallback while the flagged message is still in the thread, and editing your previous message before retrying often helps.

To stop the automatic hand-off, go to Settings > Capabilities (in Claude Code, Config > MODEL & OUTPUT) and turn off "Switch models when a message is flagged". A flagged request then pauses the conversation, and you choose whether to edit it or send it to another model yourself.

What a blocked request costs on the app and the API

In the Claude apps, what you pay depends on when the block happens. A request stopped before Fable 5 writes anything is charged at Opus rates only. One stopped partway through is charged at Fable 5 rates for the input and the text already streamed, and at Opus rates for the rest.

The Claude API does not switch models on its own. Until a developer configures fallbacks, a blocked request returns a 200 response with stop_reason: "refusal", so the integration has to check that field to notice the block. On the API, a refusal before any output is not billed, and a mid-stream one bills the input and the output already streamed.