Claude Fable 5 Opens with Guardrails

Anthropic released Claude Fable 5, a public Mythos-class model with guardrails for cyber, biology, chemistry, and distillation requests.

1. Frame 1Anthropic launches the Claude Fable 5 model for public AI users, with guardrails blocking high-risk cyber and biology requests.
2. Frame 2A gate diagram splits Fable from Mythos 5: public users get more limits, approved partners get more capability.
3. Frame 3A data path sends cybersecurity, biology, chemistry, and distillation prompts through a classifier before blocks or Opus 4.8 reroutes.
4. Frame 4The access ledger changes June 23, when subscription inclusion gives way to metered usage and higher costs.
5. Frame 5Anthropic’s benchmark says 95 percent of sessions stayed on Fable, but edge cases test the guardrails.
6. Frame 6Watch the failure mode: cautious blocks could frustrate benign users while missed blocks expose dangerous tasks.