Anthropic switched Claude Fable 5 back on: inside the 19-day suspension and the redeployment
For 19 days in June 2026, Anthropic's most capable widely released model was switched off for every user by US export controls. On 1 July, Claude Fable 5 and Mythos 5 came back with new safeguards. The exact timeline from Anthropic's own post, why it was pulled, what changed, how access worked, and what happened since, up to Fable 5.1.
In short
- Released on 9 June 2026, Claude Fable 5 and Mythos 5 were suspended for all users on 12 June, after US export controls restricted access for foreign nationals and Anthropic could not verify nationality in real time.
- The trigger, per Anthropic: Amazon researchers bypassed Fable 5's safeguards to identify software vulnerabilities and get exploit demonstration code. Anthropic says GPT-5.5, Opus 4.8 and Kimi K2.7 could do the same.
- The controls were lifted on 30 June and access restored on 1 July, with a classifier blocking the technique in over 99% of cases, a fallback to Opus 4.8, a wider safety margin, and safeguards CAISI testers called extraordinarily strong, per Anthropic.
- Since then: Fable 5.1 and Mythos 5.1 on 1 September, with fewer false positives; Mythos-class access limited to vetted US organisations for now; Fable 5 is a legacy model.
On 30 June 2026, Anthropic announced that the US export controls on Claude Fable 5 and Mythos 5 had been lifted, and on 1 July it restored access. For the previous 19 days, its most capable widely released model had been switched off for every user, after Amazon researchers found a way to make it surface software vulnerabilities. It remains one of the clearest public examples of a government pausing, then allowing, a frontier AI model.
Because the subject is sensitive, this article sticks to the primary source, Anthropic's own post, and says so when something comes from elsewhere. It was first published on 1 July 2026 and revised on 17 October 2026: claims we could not trace to a source have been removed, and the last section covers what happened since.
The timeline: 19 days offline
| Date | Event, per Anthropic |
|---|---|
| Tuesday 9 June 2026 | Claude Fable 5 and Mythos 5 are released. |
| Friday 12 June 2026 | The US government applies export controls restricting access for foreign nationals, inside or outside the United States, effective immediately. Anthropic suspends both models for everyone. |
| 26 June 2026 | After the US government's approval, Mythos 5 access is restored for a set of US organisations. |
| 30 June 2026 | The export controls on Fable 5 and Mythos 5 are lifted. |
| 1 July 2026 | Access to Fable 5 and Mythos 5 is restored. |
Why it was pulled
The mechanism is unusual. The order restricted access for any foreign national, whether inside or outside the United States. Anthropic writes that it had "no reliable way to verify nationality in real-time", so it suspended both models for every user. Its other Claude models were not affected.
The trigger, per Anthropic, was a report in which Amazon researchers had found a method of bypassing Fable 5's safeguards, by prompting the model to identify software vulnerabilities; in that work, the model "produced code demonstrating how the relevant vulnerability could be exploited". Anthropic pushes back on the severity: it says many less capable models, including Claude Opus 4.8, GPT-5.5 and Kimi K2.7, could identify the same vulnerabilities and produce the same demonstration, and that the technique "did not expose any unique Mythos-level cyber capabilities". A frontier lab and a government reading the same demonstration very differently: that disagreement is the heart of the story.
The nuance matters. Fable 5 was not pulled for a hallucination or a content-policy slip. It was pulled because a government chose to treat a cybersecurity capability as something whose access can be export-controlled, applying export-control law to access to a model. That is a new category of event.
What changed for the redeployment
- A stronger classifier. A new classifier blocks the technique from the Amazon report in "over 99% of cases". A blocked request is rerouted to Claude Opus 4.8, and the user is told.
- A wider safety margin. Anthropic says the margin is "much larger than in any prior launch", which means more false positives on routine coding and debugging.
- Government evaluation. Testers at CAISI, the Center for AI Standards and Innovation, tested the safeguards and, per Anthropic, "agree that they are extraordinarily strong". This is a company-relayed assessment, not a published CAISI report. Anthropic also commits to giving government partners expanded early access to models and safeguards.
- An industry framework and a HackerOne programme. Anthropic is drafting, with Amazon, Microsoft, Google and other Glasswing partners, a consensus framework that scores jailbreaks on capability gain, breadth, ease of weaponisation and discoverability, and opens a HackerOne programme for cyber jailbreaks found in Fable 5.
The trade-off is stated openly: a wider margin means Fable 5 refuses more legitimate work. For the model at the top of Anthropic's public line-up, that is a deliberate choice of caution over convenience, in direct response to what got it pulled.
How access worked on 1 July
The redeployment was metered. Fable 5 came back to users globally on the Claude Platform, Claude.ai, Claude Code and Claude Cowork. For Pro, Max, Team and select Enterprise plans, it was available up to 50% of weekly usage limits through 7 July, then through usage credits; standard Enterprise had no included allowance. AWS, Google Cloud and Microsoft Foundry access was to return "as quickly as possible". Mythos 5 stayed limited to US organisations, with broader Glasswing access still being coordinated.
What happened since
- 1 September 2026: Fable 5.1 and Mythos 5.1. Same base price, cache reads four times cheaper, and, per Anthropic, about 60% fewer false positives on cyber safeguards; Fable 5.1 can help find software vulnerabilities but cannot build exploits, and dual-use cyber work such as pentesting is routed to Opus models (Anthropic, September 2026). Fable 5.1 is available on all platforms, including AWS, Google Cloud and Azure.
- Mythos-class access. Mythos 5.1 is limited to a set of US organisations for now, through verification programmes such as the Cyber Verification Program and the Life Sciences Verification Program, and Anthropic says it is coordinating with the US government to widen access to domestic and international partners.
- Fable 5 is now a legacy model, still available, with retirement not sooner than 9 June 2027 (Claude documentation).
Our read: the precedent is the story
What follows is our analysis, not Anthropic's statement.
The model coming back is not the interesting part. The interesting part is that the full loop ran in public: a lab ships a frontier model, an outside team demonstrates a capability that worries a government, export controls take it offline for everyone, the lab hardens it and coordinates with the state, and the model returns under conditions, all in under three weeks. That is, in miniature, the governance model many people expect for frontier AI.
For anyone building on these models, the lesson is concrete: never wire a product to a single model. If a workflow depended only on Fable 5 between 12 June and 1 July, it was down for 19 days through no fault of its own. Keep the model behind your own interface, with a tested path to another model, Opus, Sonnet or an open-weights option, that takes a configuration change rather than a rebuild. That is how we build our own products. For the launch itself, read what Fable 5 and Mythos 5 are; for costs and integration, Fable 5 for business; and for the model that arrived the same month, Claude Sonnet 5.
Frequently asked questions
Why was Claude Fable 5 suspended?
According to Anthropic, Amazon researchers found a way to bypass Fable 5's safeguards by prompting it to identify software vulnerabilities, and the model produced code demonstrating an exploit. On 12 June 2026, the US government applied export controls restricting access for foreign nationals; unable to verify nationality in real time, Anthropic suspended Fable 5 and Mythos 5 for everyone.
How long was Claude Fable 5 offline?
About 19 days: suspended on 12 June 2026, export controls lifted on 30 June, access restored on 1 July. Mythos 5 access for a set of US organisations was restored earlier, after the government's approval on 26 June.
What safeguards did Anthropic add to bring Fable 5 back?
A new classifier that blocks the reported technique in over 99% of cases and reroutes blocked requests to Claude Opus 4.8 with a notice, a safety margin much larger than in any prior launch, safeguards tested by CAISI, expanded early access for government partners, an industry framework for scoring jailbreaks with Amazon, Microsoft and Google, and a HackerOne programme.
Can I use Claude Fable 5 now?
Yes, as a legacy model, with retirement not before 9 June 2027. Its successor Claude Fable 5.1, released on 1 September 2026, is available on all platforms including AWS, Google Cloud and Azure. Mythos 5.1 is limited to a set of US organisations through Anthropic's verification programmes.
What does the Fable 5 suspension mean for businesses using AI?
That a frontier model can be switched off by policy, not only by an outage. Any product wired to Fable 5 alone was down for 19 days. Treat the model as a replaceable component behind your own interface, with a tested fallback to another model that only takes a configuration change.
Read next
Is your project decided?
Present it in writing. We read everything, then we answer.
Pitch a project