The Information Machine

This may be an extreme case but it still shows how quickly Fable 5 classifiers can reroute routine coding to Opus.

Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-07-02

A user reported a $321 coding session in which Anthropic's Claude Fable 5 safety classifiers repeatedly misidentified routine coding prompts as cybersecurity risks, routing 75% of the work to the more expensive Opus 4.8 model without the user's awareness.

Open original ↗

Appears in

Extraction

Topics: anthropicclaude-fable-5ai-classifiersmodel-routingai-pricing

Claims

  • Fable 5's safety classifiers can misidentify routine coding tasks as cybersecurity risks, triggering automatic fallback routing to the more expensive Opus 4.8 model.
  • In the reported case, 75% of a $321 session was handled by Opus 4.8 at $242, while Fable 5 only performed $78 of work.
  • Anthropic described task fallbacks as affecting a small fraction of sessions, but at least some users are experiencing much higher fallback rates.
  • The classifier over-triggering creates a transparency and cost-predictability problem for users who pay to use a specific model.

Key quotes

This may be an extreme case but it still shows how quickly Fable 5 classifiers can reroute routine coding to Opus.
75% of the session got routed to Opus because the new classifiers kept flagging routine coding work as cybersecurity risk.
Anthropic said a small fraction of tasks would fall back. My receipts say otherwise.