USA re-releases Claude Fable 5 and Mythos 5

This article may contain affiliate links. Learn more on our Partner Program page.

Make Apfelpatient a preferred source One click – and you'll see us more often on Google

After nearly three weeks of enforced silence, Anthropic is allowed to reactivate its most powerful AI models. The US Department of Commerce has lifted the export ban on Claude Fable 5 and Mythos 5 – the end of a conflict that was at one point discussed as a “digital kill switch” for commercial AI.

On June 9, Anthropic publicly released the Claude Fable 5, the first model in its high-performance Mythos class, accompanied by its more restricted sibling, the Mythos 5, which remained available only to select partners through the Project Glasswing program. Just three days later, a U.S. government export control directive forced the company to shut down both models worldwide. Since then, Anthropic has been negotiating with the White House - initially with partial success, but now with a complete reversal: On Tuesday evening, the Commerce Department lifted the restrictions on the Fable 5 and Mythos 5. Anthropic publicly confirmed the release shortly thereafter and announced that it would gradually restore access starting July 1.

What Anthropic has confirmed

In a post on X, the company explained that it had received notification that the Department of Commerce had lifted export controls on Claude Fable 5 and Mythos 5. Access would be restored the following day; a more detailed update would follow. Anthropic thanked its customers for their patience and everyone who had contributed to the re-availability.

The company left two crucial questions open: whether Fable 5 will be fully available to users worldwide again, and whether the return will be all at once or in stages. Further details are expected to follow shortly via the company's newsroom.

How the shutdown came about

The directive reached Anthropic on June 12 at 5:21 p.m. local time on the US East Coast. It prohibited all access to both models for foreign nationals - both within and outside the US, and explicitly also for foreign Anthropic employees. The company reportedly had only about an hour to implement it: Since a separation based on nationality could not be reliably enforced during normal operations, Anthropic decided to completely disable Fable 5 and Mythos 5 for all customers worldwide. Other models, such as Claude Opus 4.8, Sonnet 4.6, and Haiku 4.5, were spared the ban.

According to several reports, the trigger was an alleged jailbreak designed to circumvent Fable 5's security mechanisms. The Wall Street Journal reported that Amazon researchers had demonstrated that the model could be tricked into providing information that could be used for cyberattacks through a specific sequence of inputs. Discussions between Amazon CEO Andy Jassy and the White House are said to have contributed to the export order. Anthropic strongly disagreed with this assessment: The jailbreak in question was very limited and essentially amounted to allowing the model to read a codebase and fix software bugs. An examination of a demonstration revealed only a few known, minor vulnerabilities that other publicly available models, such as OpenAI's GPT-5.5, could also detect. In the company's view, such a minor finding did not justify recalling a commercially deployed model.

A two-stage compromise

The return didn't happen all at once, but in stages. In the week before the full release, Anthropic initially received permission to distribute Mythos 5 to over 100 US institutions – including companies and government agencies, exclusively within the US. Fable 5 was excluded from this initial easing of restrictions.

On Anthropic's side, co-founder Tom Brown recently took over the leadership of the negotiations, replacing CEO Dario Amodei, who had repeatedly come under government scrutiny due to his strong stance on AI security and his political positions. The corresponding letter from the Department of Commerce was most recently addressed to Brown.

The conditions of release

Commerce Secretary Howard Lutnick specified concrete conditions for the complete lifting of the export license. His letter stated that Anthropic no longer required an export license after the company committed to proactively identifying and addressing safety risks in its models, working with the government to develop protocols for future model releases, and reporting any detected malicious activity.

Lutnick subsequently confirmed the agreement himself on X. He stated that they had worked closely with Anthropic over the past two weeks to analyze and release Fable 5 – however, he did not elaborate on when and how exactly the model would return.

What this case shows beyond a single model

This episode is considered one of the most aggressive interventions to date in a commercially deployed AI model. For the first time, a government has taken a publicly available system offline via export control – an action quickly understood within the industry as a precedent for a regulatory “kill switch.” At the same time, the legal framework remained unclear: The role the US government intends to play in reviewing frontier models before their release remains open and is currently decided on a case-by-case basis.

The move was accompanied by sharp criticism from parts of the tech industry. Several executives and investors warned that the blockade would give Chinese providers of open-source models valuable time in the race for market leadership – while they are already rapidly catching up. For Anthropic, the release brings an end to a grueling dispute, but the fundamental question remains: How much control should the state claim over market access to the most powerful AI systems – and according to which rules? (Image: Shutterstock / RixAiArt)