Anthropic has announced the public release of its Mythos model, which is accessible through Claude Fable, though this version notably lacks cybersecurity capabilities. This rollout marks a shift from the previous restriction of Mythos-class models, as Anthropic aims to enhance access to its frontier AI models while still focusing on safety and risk management, particularly in the sensitive domain of cybersecurity. The omission of advanced cybersecurity tools reflects the company’s ongoing efforts to develop appropriate safeguards for its models.

Mythos: Mythos is Anthropic’s frontier AI model class, with Claude Mythos Preview demonstrating exceptional performance in cybersecurity tasks such as vulnerability identification and exploitation. Previously limited to defensive partners through Project Glasswing due to its advanced capabilities, a stripped-down public version is now being made available without those cybersecurity features. The model underpins newer releases like Claude Fable aimed at general users.
Anthropic: Anthropic is an AI research company that develops advanced frontier models including the Claude family. It recently released a public version of its Mythos-class capabilities in the form of Claude Fable after initially restricting the full Mythos model due to its powerful cybersecurity features. The company is using the rollout to balance broader access with ongoing safety considerations around the model’s dual-use potential.

Model Release: Anthropic has transitioned from restricting Mythos-class models to offering a public version through Claude Fable that retains high performance while omitting cybersecurity tools.
Safety Approach: The public Mythos rollout excludes the advanced cyber capabilities that led to initial limited access, reflecting ongoing development of safeguards for full model classes.
Industry Context: Anthropic continues to expand access to frontier models while addressing risks associated with powerful AI in sensitive domains like cybersecurity.