Anthropic blocks access to Mythos and Fable from Claude

• Anthropic has blocked the ability to generate prompts for Claude Mythos 5 and Claude Fable 5 that could be used to bypass restrictions for the United States
• Fable 5 contains a filter that prevents prompts for accessing the tool for certain critical users
• Mythos 5 has been restricted from being used by external organizations in the United States, while Fable 5 will have to follow the same restrictions via specific safety mechanisms

Anthropic said that the company’s restrictions are aimed at preventing the most dangerous capabilities of Claude Mythos 5 and Claude Fable 5, and that the restrictions will be rolled out starting 1 February 2026. The company says the changes will be applied to all users.

In a statement, the company noted that Fable 5 has “proven” safeguards against certain types of jailbreaks. That said, Anthropic acknowledged that even with the existing standard model safety measures, there may be cases where the model could still be tricked into following instructions that it shouldn’t comply with.

For that reason, Fable 5 will have a public-facing version of the model, while the internal version will be subject to additional safeguards. A win-win situation for both sides: the model will be available to users, and Anthropic will be able to ensure that it’s used safely.

After the release of Fable 5, Anthropic said it would implement a set of restrictions and would also continue to improve the model’s safety. In particular, the company will deploy a new “jailbreaks” technique to prevent the model from complying with instructions that could be used to bypass its safeguards.

In addition, with the new materials, Mythos 5 will be available for “trusted” U.S. organizations within the Project Glasswing framework, while Fable 5 will be limited to a smaller set of users who must pass an audit. At the same time, it is expected that access to Fable 5 for individual users may be limited until 7 February 2026.

The main reason for the restrictions is that the company believes that the model’s capabilities could be used by bad actors to create dangerous content. Anthropic says it has identified vulnerabilities in Fable 5 that could be exploited to bypass the model’s safety measures, and that it is taking steps to address them.

Tags: USA/Politics/Technology/AI