In short: yes, Claude Fable 5 is back—and this time for everyone. On June 30, 2026, the US Department of Commerce lifted the export controls that had applied to Fable 5 and its twin Mythos 5 since June 12. Anthropic announced the model’s return globally starting Wednesday, July 1, on the Claude Platform (API), claude.ai, Claude Code and Claude Cowork. The key point for non-US users: this return is worldwide, not restricted to US accounts—it’s now Mythos 5 that stays confined to around a hundred “trusted” US organizations.
The redeployment does come with new conditions, though: a new cybersecurity classifier (some tasks, including routine coding and debugging, fall back to Opus 4.8), temporary usage limits, and 30-day retention of data for Mythos-class models. Full details below.
If you landed here searching for “Claude Fable return”, “Claude Fable available” or “is Fable back”, here’s the clear answer, the full timeline, the reasons for the shutdown and what concretely changes with this redeployment.
Is Claude Fable back today?
Yes. As of July 1, 2026, Claude Fable 5 is once again available to users worldwide on Anthropic’s own surfaces: the Claude Platform (API), claude.ai, Claude Code and Claude Cowork. It’s the outcome of two and a half weeks of negotiations between Anthropic and the US administration.
Two practical caveats. First, the rollout may be gradual: depending on your time zone, the model may only appear in the selector later in the day. Second, access via the partner cloud platforms—AWS (Bedrock), Google Cloud (Vertex) and Microsoft Foundry—is not yet restored: Anthropic says it will re-enable them “as quickly as possible.” If you call Fable 5 through one of these hyperscalers, you’ll need to wait a little longer.
What changes with this redeployment
Fable 5 isn’t coming back unchanged. Anthropic adjusted several parameters to get the government’s green light.
A new cyber classifier (and fallbacks to Opus 4.8)
The core of the setup: a safety classifier trained specifically to detect and block the bypass technique reported by Amazon. Anthropic says it intercepts that technique in more than 99% of cases. In practice, when a request is deemed sensitive from a cybersecurity standpoint, Fable 5 does not answer itself: the task is routed to Opus 4.8 and the user is informed. Anthropic acknowledges the downside: more false positives on perfectly legitimate coding and debugging tasks, which will also fall back to Opus 4.8 in the near term. The company says it will refine these classifiers over the coming weeks to better distinguish genuine misuse from legitimate requests.
Temporary usage limits
For Pro, Max, Team and select Enterprise plans, Fable 5 is included at up to 50% of weekly usage limits through July 7; beyond that, access is via a usage credits system. Standard Enterprise seats have no included Fable 5 allowance—access goes directly through usage credits.
30-day data retention for Mythos-class models
Anthropic now requires 30-day retention of all traffic on Mythos-class models (Fable 5, Mythos 5 and future models of equal or higher capability), including on third-party surfaces. This data is not used to train new models or for any non-safety purpose; it’s meant to detect complex attacks and new jailbreaks and to reduce false positives. The company says it has added safeguards: logging all human access and deleting the data after 30 days in almost all cases.
A jailbreak-severity framework and more government collaboration
In parallel, Anthropic has begun drafting—together with Amazon, Microsoft, Google and other Glasswing partners—a consensus framework for assessing the severity of AI jailbreaks and how developers should respond, and it invites other industry players to join. The company is also scaling up its collaboration with the US government on testing and safeguards: pre-release access to models for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.
Why Claude Fable had been suspended: the real reasons
A useful reminder, because the episode is unprecedented: it was neither an outage nor a bug, but a regulatory decision.
The full timeline
- June 9, 2026: Anthropic opens Claude Fable 5 to the public—its most powerful model ever made widely available. Its twin Mythos 5, fundamentally identical but with fewer safeguards, stays reserved for the closed Project Glasswing program.
- June 12, 2026, 5:21 PM ET: Commerce Secretary Howard Lutnick sends Anthropic an export-control directive, grounded in national security.
- That same evening: Anthropic disables Fable 5 and Mythos 5 for all its customers.
- June 26, 2026: first thaw—the government authorizes redeploying Mythos 5 to around a hundred “trusted” US organizations (federal agencies and critical-infrastructure operators). Fable 5 remains excluded from this authorization.
- June 30, 2026: the Department of Commerce lifts the export controls on both Fable 5 and Mythos 5. Anthropic announces access will be restored the next day.
- July 1, 2026: Fable 5 is available again worldwide on claude.ai, the Claude Platform, Claude Code and Claude Cowork. Mythos 5 stays reserved for approved US organizations.
The reason the government cited
The June 12 directive ordered access to Fable 5 and Mythos 5 to be blocked for any foreign national, inside or outside the United States—including Anthropic’s own foreign-national employees. The technical trigger: a report by Amazon researchers describing a “jailbreak” method that got Fable 5 to identify software vulnerabilities—and, in one case, to produce code showing how to exploit one of them. Once it reached the authorities, the report raised concerns about the offensive cyber capabilities of public LLMs.
Anthropic’s position: a misunderstanding, now settled by testing
Anthropic always disputed the severity of the issue—and its testing, conducted with the government and Amazon over two weeks, largely bears it out. According to the company, less capable models—including Opus 4.8, GPT-5.5 and Kimi K2.7—identified the same vulnerabilities as Fable 5 in the report; and for the exploit demonstration, every model tested produced the same result. In other words, this was routine defensive security work, with no offensive capability unique to Fable 5. That finding, combined with the new classifier, is what broke the deadlock.
Why was everyone cut off, not just foreign nationals?
Because there is no reliable way to verify nationality on every API request, across dozens of cloud platforms. To stay compliant, Anthropic chose to shut everything down rather than risk a violation—turning a measure targeting foreign nationals into a total global block.
And Mythos 5? Still under lock and key
Don’t confuse the two models. While Fable 5 returns for everyone, Mythos 5 stays reserved for a closed list of around 100 “trusted” US organizations (federal agencies, critical-infrastructure operators), authorized since June 26. Anthropic is continuing talks with the government to widen this access to the other—domestic and international—partners in the Glasswing program. Mythos 5 is the same base model as Fable 5 but without the cyber safeguards: it’s neither open to the public nor accessible to foreign users.
Claude Fable return outside the US: what can non-US users do?
This is the point that changed most—and a chance to correct a prediction. Many feared that countries targeted by the measure (as “foreign nationals”) would be left out of a first return reserved for US accounts. That’s not what happened: the lifting of export controls is complete, and Fable 5 returns worldwide. Users in France, the UK or Canada therefore have access on the same footing as US users, on claude.ai, the API, Claude Code and Claude Cowork (subject to the gradual rollout mentioned above). Only Mythos 5 remains out of reach.
The affair kept a strong Franco-American dimension. The G7 summit was held in Évian-les-Bains, France: Donald Trump met Dario Amodei there, and the White House indicated the president had “eased” his national-security concerns after that exchange. Emmanuel Macron, for his part, had called the US AI export controls “nationalistic.” The swift resolution shows the matter was playing out as much on diplomatic ground as on technical ground.
“I see Fable in the app but it doesn’t work”
During the suspension, the Fable 5 name stayed visible in the model selector without being active—a mere interface artifact. Since July 1, the model should work; if you still get an error or a silent fallback to Opus 4.8, two likely explanations: the gradual rollout (wait a few hours, especially outside the US) or a call via a cloud platform (AWS/GCP/Foundry) not yet restored. Note too: a fallback to Opus 4.8 can now be deliberate, triggered by the new cyber classifier on a request judged sensitive.
Beware fake “unlock” methods
With the block now lifted, the question is moot. For the record, during the suspension no “trick” (miracle config file, secret prompt, alternative build, unofficial mirror) worked: the block was enforced server-side, by government decision. The pseudo-mirrors and “unlocked versions” didn’t offer the real Fable 5 and mainly exposed users to concrete security risks—a caution worth keeping for next time.
The real lesson: make the model layer interchangeable
Even with Fable back, this episode leaves a lesson that goes beyond Anthropic: a frontier model can vanish overnight for a non-technical reason—regulatory or geopolitical. For anyone building an online store or a module that relies on AI, the design rule is clear: the model layer must be interchangeable. If you can’t switch your traffic from one model to another through simple configuration, without redeploying, your abstraction needs rethinking. Fable 5’s own behavior illustrates it: when its classifier blocks a request, it falls back to Opus 4.8 on its own. It’s also why multi-provider alternatives (open-weight models like GLM 5.2, or other vendors’ models) keep drawing interest. The good news: all the other Claude models—Opus 4.8, Sonnet 4.6, Haiku 4.5—stayed available throughout the crisis and still are.
FAQ — Claude Fable return
Is Claude Fable 5 back?
Yes. As of July 1, 2026, following the US Department of Commerce lifting the export controls on June 30. The model is again available worldwide on claude.ai, the Claude Platform, Claude Code and Claude Cowork.
Does the Fable 5 return apply outside the US?
Yes. Unlike what the initial Mythos 5 thaw suggested, the Fable 5 return is worldwide: non-US users have access like everywhere else. Only the rollout may be gradual depending on time zone.
Can I use Fable 5 on AWS, Google Cloud or Microsoft Foundry?
Not yet. The return covers Anthropic’s own surfaces first (claude.ai, API, Claude Code, Claude Cowork). Anthropic says it will restore access on AWS Bedrock, Google Vertex and Microsoft Foundry “as quickly as possible.”
Why do some of my requests fall back to Opus 4.8?
Fable 5 returns with a new cyber classifier. When a request is judged sensitive (or, in the near term, for some routine coding/debugging out of caution), the task is routed to Opus 4.8 and you’re informed. Anthropic is set to refine these classifiers over the coming weeks to reduce false positives.
Are there usage limits?
Yes, temporarily. For Pro, Max, Team and select Enterprise plans, Fable 5 is included up to 50% of weekly limits through July 7, then via usage credits.
What’s the difference between Fable 5 and Mythos 5?
It’s the same base model. Fable 5 ships with safety classifiers that may decline certain requests (cybersecurity, biology, chemistry…) and fall back to Opus 4.8. Mythos 5 has fewer and stays reserved for hand-picked US organizations (Project Glasswing, then the June 26 “trusted” list)—it’s not open to the public.
Is my data retained?
For Mythos-class models (including Fable 5), Anthropic now applies 30-day retention of all traffic, for safety purposes only (detecting attacks and jailbreaks, reducing false positives). This data is not used to train the models.
Article updated on July 1, 2026, after the export controls were lifted and Anthropic’s official announcement. Sources: Anthropic’s official blog and account, CNBC, VentureBeat, The Hacker News.