Anthropic has unveiled Claude Fable 5, an artificial intelligence (AI) model designed for general users that rivals its Mythos level capabilities. The company has also implemented robust safeguards to prevent misuse in sensitive areas like cybersecurity.
On Tuesday, Anthropic announced the release of Claude Fable 5, a consumer-friendly version of its Mythos-grade AI model, alongside Claude Mythos 5, a model specialized for cybersecurity.
Fable 5 represents Anthropic’s most advanced model to date. The company claims it excels across nearly all performance benchmarks and has demonstrated exceptional capabilities in fields such as software engineering, knowledge work, and scientific research.
While Fable 5 shares the same underlying architecture as Mythos 5, it incorporates additional safety measures to prevent misuse.
Previously, Anthropic had limited the preview of its first Mythos model to select companies through its security consortium, Project Glasswing, in April. However, with the development of powerful new safeguards, the company can now offer a Mythos-grade AI model to the general public.
Anthropic stated that they have been refining these safety measures over recent months, carefully calibrating them to ensure a secure launch of the model.
The company added that while some benign requests might be blocked, initial data shows that over 95% of sessions proceed without interruption.
These safety measures extend beyond cybersecurity, also covering queries related to biological and chemical threats that could be weaponized, as well as suspected attempts at unauthorized distillation of the AI model’s capabilities.
The unrestricted Mythos 5 will be distributed to a select group of vetted organizations through Project Glasswing, a collaboration between Anthropic and the U.S. government.
This means that recent Project Glasswing members, including South Korean firms like Samsung Electronics, SK Hynix, and SK Telecom, as well as the Korea Internet & Security Agency (KISA), are likely to gain access to Mythos 5.
Anthropic has implemented a data policy that retains information generated by both models for 30 days. This data will be used to prevent new attacks and identify false positives.
The new models have shown improved performance metrics compared to the Mythos preview released two months ago. In the ExploitBench assessment of cybersecurity capabilities, Mythos 5 achieved a score of 78%, surpassing OpenAI’s GPT-5.5 (34%), Anthropic’s own Opus 4.8 (40%), and the previous Mythos preview (69%).
However, Anthropic noted that Fable 5’s performance couldn’t be measured due to its integrated safety mechanisms.
Fable 5 is available for use starting Tuesday. Until June 22, it will be offered to current paid subscribers at no extra cost, after which a separate fee will be required. Anthropic plans to reintegrate Fable 5 into existing subscriptions once sufficient server capacity is secured.