- → THE June 30, 2026the US government has lifted export controls on Claude Fable 5 and Mythos 5.
- → Fable 5 is back accessible worldwide since July 1 : Claude.ai, Claude Code, Claude Cowork and theAPIApplication Programming Interface, the programmatic access channel to a service or model..
- → Anthropic has led to a enhanced security classifier which blocks the circumvention technique in more than 99% of cases.
- → The flaw came from a test of Amazon researchers ; Anthropic’s most powerful model takes its place again.
Nineteen days of suspension, and then the outcome. After being disconnected by order of the American government on June 12, Claude Fable 5, Anthropic’s most powerful model, is once again accessible. The United States lifted its restrictions on June 30, and the model returned online on July 1. Epilogue to one of the strangest standoffs of the year between an AI company and its own state.
We told about the surprise suspension of Fable 5; here’s how the story ends, and what it leaves behind.
June 30 · lifting of restrictions
July 1 · back online
19 days of suspension
99% · bypasses blocked
01, The return
Fable 5 once again accessible everywhere
On June 30, 2026, the American administration lifted the export control directive which weighed on Claude Fable 5 and its Mythos 5 variant. The next day, July 1, Anthropic redeployed the model for all users in the world: on Claude.ai, in Claude Code, in the new Claude Cowork space and via the API.
For Pro, Max, Team, and select Enterprise plans, Fable 5 is included with up to 50% of weekly usage limits until July 7, before switching to usage credits. In other words, the flagship model takes up exactly the place it occupied before the cut.
02, The origin
A flaw found by Amazon researchers
We now know more about what triggered the crisis. The directive followed a report in which Amazon researchers found a method to bypass Fable 5’s safeguards by causing it, through a series of queries, to identify several software vulnerabilities.
It is this risk, that of an AI capable of helping to identify cybersecurity vulnerabilities, which had justified the emergency measure. Anthropic had judged the flaw to be minor and already known elsewhere; the government preferred to cut first and then discuss.
03, The Fix
A reinforced safeguard to reassure
To get the green light, Anthropic did not sit idly by. The company trained an enhanced security classifier, specifically dedicated to the reported behavior, which now blocks the evasion technique in more than 99% of cases.
This is the counterpart of the return: the model returns, but with additional protection targeted at the vulnerability that started it all. A way for Anthropic to give assurances without recognizing that the model was really dangerous.
04, What’s Left
A precedent that does not disappear
The episode is closed, but it leaves a mark. For the first time, a state used export controls to pull the plug, overnight, on an already launched consumer AI product. The fact that the restriction was lifted in three weeks does not erase the precedent: governments now have a stop button, and they have shown that they can activate it.
For businesses and developers, the lesson is concrete: the availability of a cutting-edge model can depend on a political decision, not just a publisher’s choice. This is one more argument in favor of diversification, and a reminder of why the question of sovereignty, raised in Europe by actors like Mistral, is not abstract.
05, And now
Which model Claude should use today
The situation has returned to normal: Claude Fable 5 is once again Anthropic’s most capable model, available to everyone. During the suspension, Claude Opus 4.8 had taken over the role of default model, and it remains an excellent option, alongside Sonnet and Haiku, for those who want a good cost-performance ratio. Our Claude model sheet reflects the return of Fable 5.
There remains an unknown factor: if such a suspension could happen once, it could happen again. The return of Fable 5 is reassuring in the short term, but the real novelty of 2026 is that the power of a model and its availability have become two distinct things.
To go further: The suspension of Fable 5, our article · Claude Fable 5: the general public launch · The Claude model sheet · Compare the AI models
Frequently asked questions
- Is Claude Fable 5 available again?
- Yes. The US government lifted export controls on June 30, 2026, and Anthropic redeployed Fable 5 globally on July 1, on Claude.ai, Claude Code, Claude Cowork and the API. The Mythos 5 variant is also reinstated.
- Why was Fable 5 suspended?
- On June 12, 2026, the American government imposed export controls after a report by Amazon researchers showing a method to bypass the model’s safeguards and make it identify software vulnerabilities. Anthropic had cut off access for all while complying.
- How long did the suspension last?
- Nineteen days: from June 12 (cut) to July 1, 2026 (return online), after restrictions are lifted on June 30.
- What has changed to allow the return?
- Anthropic trained an enhanced security classifier, targeted to the reported behavior, that blocks the evasion technique more than 99% of the time. This is the technical counterpart which accompanied the lifting of restrictions.
- Is Fable 5 included in subscriptions?
- Yes. For Pro, Max, Team and some Enterprise plans, Fable 5 is included up to 50% of weekly usage limits (until July 7, 2026), then accessible via usage credits. In API, it is charged around $10 in input and $50 in output per million tokens.
- Should we fear a new suspension?
- The precedent now exists: a state has shown that it can pull the plug on an already launched consumer AI model. There is no guarantee that this will not happen again, which reinforces the interest in diversifying its tools and the argument for European sovereignty.
- Which model Claude should use now?
- Fable 5 once again becomes Anthropic’s most capable model. During suspension, Opus 4.8 served as a fallback and remains a great cost-performance option, alongside Sonnet and Haiku, depending on your needs.

