Particle.news
Download on the App Store

Anthropic Releases Claude Fable 5, a Guardrailed Public Version of Mythos

The launch tests whether strict technical and policy controls can let frontier AI capabilities be used widely without enabling misuse.

Overview

  • Anthropic announced Claude Fable 5 on Tuesday, June 9, 2026, making a Mythos‑class model available to developers through the Claude API and staged subscription access.
  • Fable 5 includes new classifier guardrails that detect high‑risk prompts in areas like cybersecurity, biology, chemistry, and model distillation and automatically route those requests to the less‑capable Claude Opus 4.8.
  • A less‑restricted Claude Mythos 5 will remain limited to vetted cyberdefenders and critical‑infrastructure partners through Project Glasswing as Anthropic works with U.S. government stakeholders on trusted access.
  • Anthropic set new commercial and safety rules: both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, and the company will retain Mythos‑class traffic for 30 days for safety monitoring while saying the data will not be used for training.
  • Anthropic says more than 1,000 hours of internal and external red‑teaming found no universal jailbreaks, early telemetry shows fallbacks occur in under about 5% of sessions, and the company warned capacity, pricing, and conservative filters may shape who can use the model and how it changes defenders’ and attackers’ work.