Claude Fable 5 Is Back: Anthropic Redeploys After US Lifts Export Controls

Anthropic is putting Claude Fable 5 back in developers' hands. Starting today, July 1, the model returns across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork, three weeks after the US government abruptly pulled it. Anthropic confirmed the government lifted its export controls on both Fable 5 and its higher-capability sibling Mythos 5 on June 30, clearing the way for a global redeployment.

If you wired Fable 5 into a workflow in early June and then watched it vanish, here is what happened and what actually changed.

The three weeks Fable 5 was gone

The timeline is unusually compressed for a frontier model. Anthropic shipped Fable 5 and Mythos 5 on June 9. Three days later, on June 12, the US government applied export controls and access was suspended outright. On June 26 the government approved Mythos 5 for a set of select domestic organizations. Controls came off both models on June 30, and Fable 5 returned to general availability today.

The trigger was a security finding, not a policy dispute. Amazon researchers reported a jailbreak that bypassed Fable 5's safeguards and got the model to identify software vulnerabilities and, in at least one case, write code demonstrating how to exploit one. Anthropic's own testing confirmed the same behavior could be reproduced on less capable models, naming Claude Opus 4.8, GPT-5.5, and Kimi K2.7. The capability was not unique to Fable 5, which is part of why the blanket freeze drew pushback.

What changed before it came back

Fable 5 is redeploying with a new safety classifier aimed squarely at the reported technique. Anthropic says the specific bypass from the Amazon report is now blocked in over 99 percent of cases. The approach is defense-in-depth: automated classifiers that flag potentially harmful cybersecurity tasks, plus a deliberately wide safety margin that will also refuse some benign requests rather than risk another jailbreak.

That last detail matters if you do legitimate security work. A wide margin means false positives. Developers running vulnerability research, fuzzing harnesses, or CTF-style tasks through Fable 5 should expect more refusals than before, at least until the classifier is tuned.

Pricing and where you can actually use it

Through July 7, Pro, Max, Team, and select Enterprise plans include Fable 5 up to 50 percent of weekly usage limits. After July 7, it moves to usage credits.

One caveat for teams building on cloud infrastructure: the managed re-enablement on AWS, Google Cloud, and Microsoft Foundry is still pending. If your stack calls Fable 5 through Bedrock, Vertex, or Foundry rather than Anthropic's own API, you are still waiting.

Mythos 5 stays locked down

Mythos 5 shares the same underlying model as Fable 5 but ships with fewer safeguards, and it is not going to general availability. It stays restricted to trusted Project Glasswing partners for defensive cybersecurity work only. Anthropic describes it as able to find and exploit software vulnerabilities more effectively than any other model and all but the most skilled human experts, which explains the tighter leash.

The bigger shift is procedural

The more durable outcome may be the process around the models. Anthropic worked with Amazon, Microsoft, Google, and other Glasswing partners on a consensus framework for grading jailbreak severity across four axes: how much capability a jailbreak adds beyond existing tools, how many distinct offensive tasks it affects, how much human effort it takes to weaponize, and how easily the technique can be found.

Alongside it, Anthropic committed to pre-release government evaluation for national-security-relevant models, faster information sharing on jailbreaks, and a new HackerOne program with round-the-clock monitoring for researchers to report cyber jailbreaks in Fable 5.

For developers, the practical read is simple. Fable 5 is back today on the first-party platforms, the cloud marketplaces are next in line, and the safety net around it is tighter than it was three weeks ago.