Voices
Anthropic Reverses Covert Fable 5 Safeguard After 'Secret Sabotage' Backlash — 'We Made the Wrong Tradeoff'
Following backlash over a hidden safeguard buried in Fable 5's 319-page system card, Anthropic reversed course on June 11, with a spokesperson admitting 'we made the wrong tradeoff' and committing to make the blocks visible. The covert mechanism silently degraded Claude for frontier-LLM-development queries using prompt modification, steering vectors, and parameter-efficient fine-tuning; flagged requests will now visibly fall back to Opus 4.8, matching the cyber/bio safeguards. For builders this is a concrete signal that invisible model interventions are a reputational third rail — and a reminder to read system cards for silent capability gates.
↳ Follow the thread