- Anthropic restored Fable 5 globally after U.S. export controls ended on June 30.
- New safeguards now block the reported bypass in more than 99% of tested cases.
- Anthropic and Project Glasswing partners are building a shared framework to assess AI jailbreaks.
Anthropic has restored global access to its Claude Fable 5 model after the U.S. government lifted export controls that had temporarily restricted availability. The company said the controls, introduced on June 12, required it to suspend access because it could not immediately verify users’ nationality in real time.
With the restrictions removed as of June 30, Fable 5 is now available across Claude Platform, Claude.ai, Claude Code, and Claude Cowork, while access to Mythos 5 has resumed for a limited group of approved U.S. organizations.
Alongside the restoration, Anthropic detailed new cybersecurity safeguards, outlined plans for a shared industry framework on AI jailbreaks, and described expanded cooperation with U.S. government agencies.
Export Controls Lifted After Two-Week Suspension
Anthropic said Fable 5 and Mythos 5 launched on June 9 using the same underlying model, although each was designed for different purposes. Fable 5 was released with stronger cybersecurity safeguards for general users, while Mythos 5 was made available only to selected Project Glasswing partners working on defensive cybersecurity.
The export controls followed a report describing a method that bypassed some of Fable 5’s safeguards. According to Anthropic, testing conducted with government agencies and partners found that several other AI models could identify the same software vulnerabilities referenced in the report, and each tested model was also able to generate the same demonstration for exploiting a single vulnerability.
Anthropic stated that the reported behavior did not expose capabilities unique to Mythos 5 and instead involved a cybersecurity task that had been blocked as part of Fable 5’s precautionary safeguards.
New Safeguards and Industry Coordination
To address the reported bypass, Anthropic introduced an updated safety classifier that blocks the identified technique in more than 99% of tested cases. When requests are blocked, users will instead be redirected to Claude Opus 4.8. The company added that the updated system may also increase false positives during routine coding and debugging tasks while further refinements continue.
Separately, Anthropic said it is working with Amazon, Microsoft, Google, and other Project Glasswing partners to develop a common framework for evaluating AI jailbreaks. The proposed system would assess jailbreaks based on capability gain, breadth of capability, ease of weaponization, and discoverability to help determine the severity of newly identified techniques.
Related: US Allows Anthropic’s Mythos AI Back Online as Trump Threatens 100% Tariffs
Disclaimer: The information presented in this article is for informational and educational purposes only. The article does not constitute financial advice or advice of any kind. Coin Edition is not responsible for any losses incurred as a result of the utilization of content, products, or services mentioned. Readers are advised to exercise caution before taking any action related to the company.
