Claude Fable 5 Returns to Accessibility: The Department of Commerce Revokes Export Controls (Also on Mythos)
On June 30, the U.S. Department of Commerce revoked the export controls on Claude Fable 5 and Mythos 5, the two flagship models of Anthropic. The company announced that Fable 5 will be available to users worldwide starting from July 1, putting an end to a situation that arose more than two weeks ago.
The controls had been imposed on June 12 with a directive prohibiting any foreign citizen, including employees of Anthropic, from accessing Fable 5 and Mythos 5, allowing 90 minutes for compliance. Lacking a reliable method to verify nationality in real-time, the company then chose to disable access for all users, including those based in the United States.
The origins of the restriction stem from a report by Amazon researchers, who identified a method to bypass the guardrails of Fable 5: bypassing the filters, the model was able to identify software vulnerabilities and, in one case, produce demonstration code to exploit them.
The New Classifier and Tests with Amazon
In the official announcement of the return, Anthropic downplays the significance of that technique: it did not expose unique offensive capabilities of Mythos level but reflected an edge case in the guardrails of Fable 5 related to routine defensive cybersecurity work. Tests conducted together with Amazon show that less capable models, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7, could identify the same vulnerabilities, and that each tested model was able to reproduce the exploitation demonstration of the only actually exploitable flaw.
Operationally, Anthropic states that it has trained an enhanced security classifier that blocks the technique described in the Amazon report in over 99% of cases; in the remaining fraction, the model may return information not detailed enough to aid an attacker. The classifier has been tested and approved by researchers from CAISI, the Center for AI Standards and Innovation within the same Department of Commerce.
The revocation came after Secretary of Commerce Howard Lutnick informed the company that an export license is no longer necessary. In return, Anthropic agreed to proactively identify and address security risks of the models, to work with the government on protocols and standards for future releases, and to report any detected malicious activities. It’s worth noting that Mythos 5 had already been restored on June 26 for a group of U.S. organizations linked to critical infrastructure and the Glasswing program.
A Shared Framework on Jailbreaks
Around the Anthropic case, the company is also building a longer-term structure. Together with Amazon, Microsoft, Google, and other Glasswing partners, the company is developing a shared framework to assess the severity of jailbreaks according to four criteria: Capability gain, Breadth of capability gain, Ease of weaponization, and Discoverability. A new program on HackerOne is also planned for reporting potential cyber jailbreaks in Fable 5, and in the coming weeks, early access to frontier models will be made available for designated government partners, in anticipation of independent pre-release assessments.
Practically, Fable 5 will be available starting July 1 on Claude Platform, Claude.ai, Claude Code, and Claude Cowork, while restoration on AWS, Google Cloud, and Microsoft Foundry is indicated as imminent. For selected Pro, Max, Team, and Enterprise plans, the model will be included up to 50% of the weekly usage limits until July 7, and subsequently accessed via usage credits.