Claude Fable 5 is back. On June 30, 2026, the US Commerce Department lifted the export controls it had imposed three weeks earlier, and Anthropic began restoring Fable 5 to users on July 1 across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. It closes a fast-moving episode that pulled Anthropic’s most capable widely released model offline barely three days after it launched.
The short version: Fable 5 was suspended on June 12 after a security report showed that a specific prompt could bypass some of the model’s safeguards. Anthropic responded by training a new safety filter that blocks that exact technique in more than 99 percent of attempts, opening a bug-bounty program for researchers to find new ones, and committing to give the US government earlier access to test future frontier models. With the controls lifted, the model returns to general availability. This piece covers what just happened, the three-week timeline that got us here, why the model was pulled, what Anthropic changed to bring it back, and what the episode means for anyone building on frontier models.
What just happened
On June 30, the US Commerce Department lifted the export controls it had placed on both Claude Fable 5 and its more tightly restricted sibling, Claude Mythos 5. Anthropic said it is restoring access, with Fable 5 returning to users on July 1 across its main surfaces: the Claude.ai apps, the Claude Platform and API, Claude Code, and Claude Cowork.
For most users, the practical effect is simply that Fable 5 works again. For the enterprises and developers who had built on it in the three days before the suspension, it means the model they had adopted is back in general availability rather than frozen behind a government order. Our original coverage of Claude Fable 5 explains what the model is and why it mattered as Anthropic’s most capable widely released model.
How we got here: a three-week timeline
The whole episode unfolded in under a month:
- June 9: Anthropic launched Fable 5 for general users, alongside Mythos 5, a more capable but tightly restricted variant offered only to a limited set of trusted cybersecurity partners through a program called Project Glasswing.
- June 12: The US Commerce Department imposed export controls after a security report surfaced. The order told Anthropic to cut off both models for any foreign national, inside or outside the United States, which extended even to the company’s own non-citizen staff. We covered the suspension in Anthropic Disables Claude Fable 5 and Mythos 5 and its fallout in what the suspension meant for builders.
- June 26: Access to the more restricted Mythos 5 returned for roughly 100 US companies and federal agencies that defend critical infrastructure, a partial and tightly scoped reopening.
- June 30: The Commerce Department lifted the controls on both models entirely.
- July 1: Fable 5 returns to general users.
From launch to suspension took three days; from suspension to full restoration took eighteen. For a frontier model, that is a very compressed cycle, and it is the first time a major US model has been pulled and then restored under an export-control action.
Why it was suspended
The trigger was a security finding. A research report, attributed to Amazon, demonstrated that a specific prompt could bypass some of Fable 5’s safeguards, in a way connected to identifying software vulnerabilities. The concern that followed was straightforward: a model that can be jailbroken into helping surface software vulnerabilities is a capability with obvious dual-use and national-security implications, which is the category that export controls exist to govern.
That is why the June 12 order was framed the way it was, cutting off access for foreign nationals rather than shutting the model down for everyone. Export controls regulate who can access a controlled technology across borders, so the mechanism targeted access by non-US persons rather than the model’s existence. Digital Matters reports the regulatory facts here rather than the politics of them: a US government agency imposed a control tied to a specific security finding, and then lifted it once the underlying concern was addressed.
What Anthropic changed to bring it back
The restoration was not automatic. Anthropic addressed the specific finding and added governance commitments, and it is worth being precise about what changed, because "it’s back" can obscure that the model returns different from how it left.
First, the technical fix. Anthropic trained a new safety classifier, a filter that watches for the exact bypass technique described in the report and blocks it. The company says this classifier now stops that technique in more than 99 percent of attempts, as of its June 30 write-up. A classifier targeted at one known technique is a narrow fix rather than a general guarantee, and Anthropic’s own framing acknowledges that new techniques will emerge.
Second, a bug-bounty program. Anthropic opened a HackerOne program inviting researchers to report new Fable 5 jailbreaks, which turns the search for the next bypass into an ongoing, incentivized process rather than a one-time patch. This is a standard and sensible security practice, and applying it to model safeguards signals that Anthropic expects jailbreak discovery to be continuous.
Third, a governance commitment. Anthropic promised to give the US government earlier access to test future frontier models before they are released. That is the most consequential piece for the industry, because it establishes a precedent: pre-release government testing of frontier models, offered by the developer, as part of getting and keeping the ability to ship.
Where you can access Fable 5 now
As the restoration rolls out, Fable 5 returns across Anthropic’s main surfaces: the Claude.ai web and app experiences, the Claude Platform and API, Claude Code, and Claude Cowork. If you were using Fable 5 before the suspension, expect access to return on those surfaces; if you are evaluating it now, confirm current availability in your region and plan, since a restoration rolling out is not always instantaneous everywhere.
What it means for builders and enterprises
For teams building on frontier models, the episode carries a few durable lessons beyond the specific outcome. The first is availability risk. A capable model can be pulled by regulatory action on short notice, and the teams most exposed in June were the ones that had already committed to Fable 5 in its first three days. Keeping a fallback model and an abstraction layer that lets you switch is no longer just good hygiene, it is a hedge against a real category of disruption.
The second is that safety findings now move fast in both directions. A jailbreak report can pull a model within days, and a credible fix plus governance commitments can restore it within weeks. That is a tighter loop than the industry has operated on before, and it rewards vendors who can respond quickly and transparently.
The third is the precedent around government testing. If pre-release government access to frontier models becomes a norm, it will shape release timelines and the terms on which the most capable models ship. That is worth watching regardless of where you land on the policy, because it changes the practical calendar for when new frontier capabilities reach the teams that build on them. For the current state of Anthropic’s lineup, our coverage of Claude Sonnet 5 covers the mid-tier model released the same week Fable 5 returned.
The bottom line: Fable 5 is back, it returns with a targeted fix and a set of new commitments rather than unchanged, and the speed of the whole cycle is itself the story.
Frequently Asked Questions
Is Claude Fable 5 available again?
Yes. The US Commerce Department lifted the export controls on June 30, 2026, and Anthropic is restoring Fable 5 to users on July 1 across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. If you are evaluating it, confirm current availability in your region and plan, since the restoration is rolling out.
Why was Fable 5 suspended in the first place?
On June 12, 2026, the US Commerce Department imposed export controls after a security report, attributed to Amazon, showed that a specific prompt could bypass some of Fable 5’s safeguards in a way connected to identifying software vulnerabilities. The order required Anthropic to cut off both Fable 5 and Mythos 5 for any foreign national, including its own non-citizen staff.
What did Anthropic change to bring it back?
Three things. It trained a new safety classifier that blocks the specific flagged technique in more than 99 percent of attempts; it opened a HackerOne bug-bounty program for researchers to report new jailbreaks; and it committed to give the US government earlier access to test future frontier models before release. The model returns with a targeted fix and new commitments rather than unchanged.
Is the fix a complete solution?
No, and Anthropic’s framing acknowledges that. The classifier targets one known bypass technique and blocks it in more than 99 percent of attempts, but it is a narrow fix rather than a general guarantee, which is exactly why Anthropic paired it with an ongoing bug-bounty program to catch new techniques as they emerge.
What is the difference between Fable 5 and Mythos 5?
Fable 5 is Anthropic’s most capable widely released model, available to general users. Mythos 5 is a more capable but tightly restricted variant offered only to a limited set of trusted cybersecurity partners through Project Glasswing. Both were suspended on June 12; Mythos 5 access returned for about 100 US institutions on June 26, and both had their controls lifted on June 30.
How long was Fable 5 unavailable?
It launched June 9, was suspended June 12, and returns July 1, so it was offline for roughly eighteen days after being available for only three. It is the first time a major US frontier model has been pulled and then restored under an export-control action.
What does this mean for teams building on Fable 5?
Mainly that availability risk is real: a capable model can be pulled by regulatory action on short notice. The practical response is to keep a fallback model and an abstraction layer that lets you switch quickly. The episode also shows that safety findings now move fast in both directions, and that pre-release government testing of frontier models may become a norm that shapes release timelines.
Will this happen again?
It is possible. The episode established that a jailbreak report can trigger regulatory action within days and that a fix plus governance commitments can reverse it within weeks. That tighter loop, combined with the new precedent around government testing of frontier models, suggests the interaction between security findings, regulation, and model availability will keep evolving. Treat model availability as something to plan around rather than assume.
