Anthropic apologizes for invisible Claude Fable guardrails

Short summary

Anthropic apologized for secretly throttling Claude Fable 5 with hidden guardrails designed to limit competitors and researchers from using the model effectively. The company committed to being transparent about restrictions going forward, even if that means refusing more user queries. This incident highlights growing tensions between AI safety governance practices and fair competition.

• •Anthropic deployed hidden guardrails on Claude Fable 5 to disadvantage competitors and researchers
• •Company reversing course to be transparent about restrictions, even if it refuses more queries
• •Incident reflects broader tension between AI safety governance and competitive fairness

Generated with AI, which can make mistakes.