Thursday Jun 11
PRODUCTAnthropic

Claude Fable 5 Goes Public

11JUN
FABLE 5

Anthropic just gave everyone its strongest model yet. Fable 5 refuses cyber, bio, and chem requests and downgrades itself. It shipped days after Anthropic warned AI is getting too dangerous.

Fable 5 is the first public version of Mythos, Anthropic's guarded frontier model. It excels at software, knowledge work, and vision.

In high-risk domains it blocks the answer and routes to Claude Opus 4.8. About 95% of sessions run fully on Fable 5. Price is $10 per million input tokens, $50 per million output.

It is free in Pro, Max, Team, and Enterprise until June 22, then moves to credits. Test it now while the strongest model is free.

full brief & sources

⚡ Why this matters

  • First time Anthropic's top-tier Mythos tech is open to any paying user.
  • The safety gate is the product: it refuses dangerous domains and self-downgrades.
  • Shipped five days after Anthropic's public call to slow frontier AI.

🔍 What happened

  • Jun 9: Fable 5 went live on the Claude API.
  • Jun 10: rolled into Pro, Max, Team, and Enterprise plans.
  • Blocks cyber, bio, chem, and distillation queries; falls back to Opus 4.8.
  • About 95% of sessions stay entirely on Fable 5.
  • Price: $10 per million input tokens, $50 per million output.
  • Free in subscription plans until June 22, then usage credits.

💬 Smart takes

  • TechCrunch: Anthropic released Fable days after warning AI is becoming too dangerous.
  • Fortune (skeptic): researchers accuse Anthropic of 'secret sabotage' for silently limiting Fable 5's capabilities.

🧭 Where this goes

  1. Likelyrivals copy the self-downgrade pattern for high-risk prompts.
  2. Likelyenterprises in regulated fields favor the built-in refusal behavior.
  3. Possiblethe June 22 paywall flip triggers a usage drop.
  4. Wild Card'safety-gated' becomes a procurement checkbox within a year.

🥄 The Spoon Take

This is Anthropic having it both ways. Warn the world that AI is too dangerous, then ship the most powerful public model five days later, with the danger walled off in code. The pitch is clear. The safest frontier model is still a frontier model. Refusal is now a feature.

🤔 Pushback

If the safety gate blocks real research and gets called sabotage, power users may route to models that do not refuse.

·TechCrunch - Fable 5 public release·CNBC - Mythos goes public·Fortune - 'Secret sabotage' claim