US Government Lifts Export Controls On Anthropic’s Claude Fable 5 And Mythos 5: What You Need To Know

Anthropic has restored access to Claude Fable 5 and Claude Mythos 5 after the US government lifted the export controls that forced both frontier models offline in mid-June. Fable 5 returns to users worldwide from today, while Mythos 5 has been reinstated for a set of approved US organisations.

The controls were imposed on June 12th, when the US government moved to restrict access to the two models for foreign nationals, whether inside or outside the United States. Because the directive took effect immediately and Anthropic had no reliable way to verify nationality in real time, the company suspended access to both models for everyone. Those restrictions were lifted on June 30th.

What triggered the suspension

Anthropic released Fable 5 and Mythos 5 on June 9th. The two share the same underlying model, but Fable 5 shipped with the strongest safeguards the company has ever applied, making it suitable for general use. Mythos 5, which carries fewer safeguards, was limited to a small group of trusted partners in Anthropic’s Project Glasswing programme for defensive cybersecurity use.

The export directive followed a report in which Amazon researchers found a way to bypass Fable 5’s safeguards, prompting the company to identify several software vulnerabilities and, in one case, produce code demonstrating how a vulnerability could be exploited. Anthropic spent the following two weeks reviewing the report with the government and partners, including Amazon.

The company’s testing concluded that the flagged behaviour was not unique to its frontier models. Less capable systems — including Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — could identify the same vulnerabilities, and every model tested could reproduce the single exploit demonstration. Anthropic characterised the technique as a borderline case that touched only routine defensive cybersecurity work, rather than any Mythos-level capability.

The fix

Working with the government, Anthropic trained an improved safety classifier that targets and blocks the specific behaviour described in the Amazon report. Users whose requests to Fable 5 are blocked will be notified, and those requests will instead be routed to Opus 4.8. The company says the new classifier stops the reported technique in more than 99% of cases, though it also flags benign coding and debugging requests more often — a trade-off Anthropic says it will keep refining to cut false positives.

Researchers from the US Department of Commerce’s Center for AI Standards and Innovation tested both the old and new safeguards and agreed they are extraordinarily strong.

How to access Fable 5 and Mythos 5

Fable 5 is available from today to users globally across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise plans, it’s included for up to 50% of weekly usage limits until July 7th, after which access continues via usage credits. Standard Enterprise seats have no included allowance and must enable usage credits to use the model at all. Anthropic says it will re-enable access on AWS, Google Cloud, and Microsoft Foundry as quickly as possible.

Mythos 5, the more powerful and less restricted model, remains tightly held. Access has been restored to a set of US organisations following government approval on June 26th, and Anthropic says it continues to coordinate with the government to expand the Glasswing programme to a broader group of domestic and international partners.

A push for industry-wide jailbreak standards

Beyond the immediate resolution, Anthropic used the episode to argue that the industry lacks a consistent way to assess the severity of AI “jailbreaks” — techniques that bypass a model’s safeguards. Together with Amazon, Microsoft, Google, and other Glasswing partners, the company is drafting a consensus framework that would score a given jailbreak on four criteria: capability gain, breadth of that gain, ease of weaponisation, and discoverability.

The company is also standing up a team for round-the-clock monitoring of jailbreak submission channels and has launched a HackerOne programme inviting security researchers to submit potential cyber jailbreaks in Fable 5 for review. Alongside this, Anthropic outlined deeper collaboration with the US government, including pre-release access for government evaluators, faster information sharing on safeguards, and dedicated compute for joint AI security research.

The events sit against the backdrop of the June 2nd Executive Order on Promoting Advanced Artificial Intelligence Innovation and Security, which Anthropic says it worked on closely with agencies across the US government over the preceding ten weeks. The company’s stated hope is that the framework and its government collaboration could form the basis for systematic, codified rules applied equally to frontier model developers and serve as a template for global coordination on AI risk.