Anthropic details unusually extensive guardrails for Fable 5

Anthropic has published more detail on the safeguards surrounding Claude Fable 5, the first publicly available version of its Mythos-class technology. The company describes the controls as the most extensive it has deployed, with restrictions focused on serious misuse in biology, chemistry, cybersecurity, and model distillation. High-risk requests can be blocked or routed to the less capable Claude Opus 4.8. Anthropic is also being explicit that the protections have weaknesses, making the release a live test of whether capability-selective guardrails can contain the most dangerous uses without removing the model's broader value. Microsoft has separately restricted Fable 5 inside its products, showing that downstream platforms may apply their own risk decisions.

Key details: June 9 release; guardrail details reported June 10, 2026, Claude Fable 5, Anthropic's most extensive safeguards to date, Restrictions target biology, chemistry, cyber, and distillation, Microsoft separately restricted access in its products.

Continue swiping for more AI Brief stories.