Anthropic announced today that its public-facing Mythos-class model, the Claude Fable 5, has reopened to users as the U.S. government lifts export restrictions previously imposed on the model. Currently, Claude users have seen the Fable 5 option in the product interface, and Anthropic is also simultaneously pushing relevant updates through in-app messages.
According to official instructions, from now until July 7, eligible Claude subscribers can use up to 50% of the weekly usage quota of their respective packages to call Fable 5; once this proportion limit is reached, continuing to use the model will require additional usage points. After July 7, Fable 5 will transition entirely to access using points.
Fable 5 is Anthropic’s first Mythos-class model available to the public, officially launching as early as June 9 this year. Anthropic said that Fable 5's comprehensive capabilities have exceeded all its previous models open to the public, and have demonstrated "excellent performance" in areas such as software engineering, knowledge work, visual understanding, and scientific research. The Fable 5 outperformed Anthropic's own Opus series models on longer, more complex tasks and was able to continue working autonomously without human intervention for longer.
When launching Fable 5, Anthropic set up relatively conservative security measures for the model to reduce the risk of abuse. However, the Trump administration subsequently issued an export control order for this model, requiring restrictions on access by foreign users. Because Anthropic was unable to effectively verify the nationality or identity information of all users, it was forced to suspend global access to Fable 5. At the same time, the company has also paused access to Mythos 5, a model of its next-generation system under Project Glasswing for large enterprises and U.S. federal agencies to help defend critical infrastructure.
This executive order stems from a test experiment by the Amazon research team: researchers found a prompt word (prompt) that can bypass Fable 5's built-in security protection, allowing the model to discover potential security holes in the software system. Anthropic subsequently launched an internal investigation and found that not only Fable 5, but its older versions and similar models launched by other companies were also able to locate the same vulnerability under similar prompts. To this end, Anthropic has launched a new classifier to block such offensive prompts, which it claims can block the use of this technology in more than 99% of scenarios.
In terms of access policies, Fable 5 is currently open to Claude Pro, Max, Team, and some enterprise-level subscription customers. Additionally, Anthropic has restored Mythos 5 access to U.S. institutions participating in “Project Glasswing.” The company stated that it will further deepen its cooperation with the US government in the future and carry out more work in model pre-release testing, information sharing and security research collaboration.