Is Claude Fable 5 Back? Yes, With New Safeguards

Claude Fable 5 status: available again since July 1

Claude Fable 5 is available again, switched back on for users everywhere on Wednesday, July 1, 2026. It had been offline for 19 days.

The model launched on Tuesday, June 9, 2026, alongside Mythos 5, and was pulled three days later. The shutdown itself, and what each plan gets today, are covered in Claude Fable 5 Unavailable: The Ban and Your Plan.

US export controls on Anthropic lifted June 30

The US export controls on Fable 5 and Mythos 5 were lifted on June 30, 2026, which cleared the way for the next day's return. An export control is a government rule on who may receive a technology.

The directive Anthropic received on Friday, June 12, 2026, cited national security authorities and ordered it to cut off every foreign national from both models, whether inside or outside the United States. That included Anthropic's own employees who are foreign nationals.

Anthropic had no reliable way to check a user's nationality in real time. To comply, it disabled both models for everyone, US nationals included. Its other models kept running.

Back on Claude.ai, Claude Code, Claude Cowork, and the Claude Platform

From July 1, Fable 5 is open to users in every country on the four products Anthropic runs directly. Anthropic gave no date for AWS, Google Cloud, and Microsoft Foundry, saying only that it would re-enable Fable 5 there as quickly as possible.

If you call Fable 5 through one of those three clouds, check that provider's own model list. Do not assume it came back on July 1 with the rest, since Anthropic described their return only as a plan.

Fable 5 vs Mythos 5: Mythos returned June 26 for some US organizations

Mythos 5 is back only for a set of US organizations, restored on June 26, while Fable 5 is the version open to everyone. The two run on the same underlying model; the difference is the safeguards.

Fable 5 shipped with strong safeguards for general use. Mythos 5 has fewer and went only to a small number of Project Glasswing partners for defensive cybersecurity work.

Both were suspended on June 12, and Mythos 5 came back first: the government approved its restoration on June 26, and Anthropic then turned it on for those US organizations. The lifting of the controls on June 30 covered both models but did not open Mythos 5 to the general public.

The new safeguards against the jailbreak Amazon reported

Fable 5 came back with a new classifier built to stop the jailbreak in the Amazon researchers' report that led to the June 12 directive. The classifier was trained with the government, and it changes how some coding requests are handled.

The Amazon researchers' jailbreak that led to the ban

Amazon researchers found a jailbreak for Fable 5, and the government acted after learning of their report. A jailbreak is a way of getting a model to do something its safeguards are meant to stop.

As it was described to Anthropic, the technique went like this: give the model a specific codebase, and ask it to read the code and fix whatever software flaws it finds. Anthropic reviewed a demonstration. The model identified a small number of vulnerabilities that were already known and minor, and in one case it wrote code demonstrating how a vulnerability could be exploited.

Anthropic then ran the same test on other models. Claude Opus 4.8, GPT-5.5, and Kimi K2.7 found the same vulnerabilities, and every model tested could produce the same exploit demonstration.

Anthropic also says it disagreed that a narrow potential jailbreak was reason enough to recall a model deployed to hundreds of millions of people. That is Anthropic's position, not a finding by the government.

Anthropic's new classifier blocks that technique in over 99% of cases

Anthropic says its new safety classifier, trained together with the government, blocks the technique from the Amazon report in over 99% of cases. A classifier is a filter that labels each request and blocks the ones it judges unsafe.

Even before this, Anthropic says it deliberately set its safety classifiers to trigger on some requests it knows are likely benign. The 99% figure is for the one reported technique, not for jailbreaks in general.

How the stricter safeguards and classifier affect coding requests

When the classifier blocks a Fable 5 request, the request is sent to Opus 4.8 instead and you are told that it was. The new classifier also flags harmless requests more often during routine coding and debugging.

The reported technique looks a lot like everyday work: read this code, fix what is broken. Our reading is that this is why ordinary requests get caught.

In practice, if a debugging session shows a notice that Opus 4.8 handled a request, that answer did not come from Fable 5. Check for the notice before you judge Fable 5's output on that task.

How the return fits the June 2 AI executive order

Anthropic presents the whole response as consistent with the June 2 executive order "Promoting Advanced Artificial Intelligence Innovation and Security". It says it worked closely with the US government as the approach in that order was developed. Everything below is Anthropic's account, not a government statement.

The last row reaches beyond Fable 5. It applies to any model that materially advances capabilities relevant to national security, not only this one.

Rules for AI companies were moving elsewhere too. In Europe, the omnibus covered in EU AI Act Omnibus: New Deadlines and Nudifier Ban moved the high-risk rules to December 2027 and August 2028.

In the US, a subpoena that a coalition of state attorneys general served on OpenAI was reported on June 12, the same day Anthropic received its directive. That case is covered in OpenAI investigation: what state AGs subpoenaed.

Where this leaves Claude users: Fable 5 has been back worldwide since July 1, 2026, with a new classifier against the reported jailbreak. As of Anthropic's July 1 update, Mythos 5 was available only to a set of US organizations.