Voices
Sam Bowman: releasing Opus 5.5 'more likely than not' reduces misalignment risk, and Nat McAleese apologizes for Opus 5
Anthropic alignment lead Sam Bowman posted that Opus 5.5 is "sufficiently safer than its predecessors that releasing it, more likely than not, reduces risks related to misalignment," and also said he is worried the field can't keep up with rising risk. The system card's automated audit ran about 4,000 investigations from roughly 1,900 seed scenarios and reports an 85% drop in boundary-circumvention attempts versus Opus 5. Anthropic researcher Nat McAleese called 5.5 "way, way, way better than Opus 5. Sorry about that model," which is an unusually direct admission that the previous release fell short.
↳ Follow the thread