Claude Fable 5 Brings Anthropic’s Restricted AI to the Public

Claude Fable 5 has officially arrived, giving the public access to technology that Anthropic previously restricted over cybersecurity concerns. The release marks a significant shift for the AI company, which spent months limiting access to Mythos-class models because of their ability to identify software vulnerabilities and perform advanced technical tasks.

Anthropic now believes it can make the technology available more broadly without exposing the most sensitive capabilities. The company has wrapped the model in additional safeguards and positioned it as the most powerful AI system it has ever released to general users.

Anthropic Changes Course on Mythos Access

Earlier this year, Anthropic introduced Mythos as a frontier AI model with capabilities that exceeded its previous public releases. Company executives highlighted its strong performance in software engineering, research, and cybersecurity-related work.

Those capabilities also triggered concerns. Anthropic warned that unrestricted access could create new risks because advanced AI systems can help identify software flaws faster than traditional methods. The company responded by limiting access to approved organizations through Project Glasswing instead of launching the model publicly.

The latest release signals a change in strategy. Rather than keeping the technology behind closed doors, Anthropic has created a version that the public can use while restricting responses in high-risk areas.

Safeguards Replace Unrestricted Capabilities

Anthropic built Claude Fable 5 on the same foundation as its Mythos technology but added controls designed to block requests involving sensitive topics.

The model can refuse certain cybersecurity, biology, and chemistry-related requests. In some situations, it redirects users to a less capable system instead of generating a response. Anthropic says these safeguards reduce the chances of misuse while preserving the model’s broader reasoning and coding abilities.

The company also claims it conducted extensive testing before the launch. Internal teams and external researchers attempted to bypass the safeguards, helping Anthropic evaluate the model’s resilience before releasing it publicly.

A New Phase in the AI Safety Debate

The launch highlights a growing challenge for AI developers. Companies want to deliver increasingly capable models, but they also face pressure from governments, security researchers, and enterprise customers concerned about misuse.

Anthropic’s decision reflects a compromise between those competing priorities. Users gain access to a significantly more capable model, while the company retains tighter control over functions that could create cybersecurity risks.

The move may also influence how other AI companies handle future releases. As frontier models become more powerful, developers will likely face similar decisions regarding public access, safeguards, and responsible deployment.

Mythos 5 Remains Behind Controlled Access

While Claude Fable 5 is now available to the public, Anthropic continues to restrict access to Mythos 5, a less constrained version of the technology.

The company plans to distribute Mythos 5 through trusted-access programs and selected partnerships rather than opening it to all users. Anthropic argues that this approach allows security professionals and approved organizations to benefit from the model’s advanced capabilities while reducing the risk of abuse.

That decision suggests Anthropic still views parts of the technology as too sensitive for unrestricted release.

Conclusion

Claude Fable 5 represents Anthropic’s attempt to balance capability and safety as AI systems continue to advance. The company has moved technology that once remained behind restricted-access programs into the public domain, but it has done so with significant safeguards in place.

The release provides a glimpse into how AI developers may handle future frontier models. Instead of choosing between full public access and complete restriction, companies may increasingly rely on layered controls that limit sensitive capabilities while allowing broader use of advanced AI systems.