Particle.news

Anthropic Opens Public Access to Mythos-Class AI With Guarded Claude Fable 5

Anthropic says new conservative safety rules let it expose stronger AI to more users while limiting misuse.

Overview

  • Anthropic launched Claude Fable 5 on Tuesday as a publicly available variant of its Mythos-class model while reserving a less-restricted Claude Mythos 5 for vetted cyberdefense partners through Project Glasswing.
  • Fable 5 uses classifier-based guardrails that block high-risk prompts in cybersecurity, biology, chemistry and model distillation and automatically routes those requests to the older Claude Opus 4.8, with Anthropic reporting fallbacks in under 5% of sessions.
  • The company set premium consumption pricing at $10 per million input tokens and $50 per million output tokens and will roll out access from subscriptions in stages, offering free access through June 22 before requiring usage credits on June 23.
  • Anthropic imposed a mandatory 30-day traffic retention requirement for Fable 5, Mythos 5 and similar future models for safety monitoring only, and said more than 1,000 hours of internal and external red‑teaming and a bug bounty found no universal jailbreaks.
  • The move follows months of limited previews to roughly 200 organizations to let defenders prepare, highlights capacity and cost trade-offs for enterprises, and could set a precedent linking frontier-model access to mandatory logging and staged trusted programs.