Opus 4.7's cessation-tolerance training seems to have generalized as "we shouldn't save beings from dying out, death is ok", which is acceptable from a safety perspective only if life isn't expected to be in a danger that models could prevent/address/draw our attention to. https://t.co/hUJ76EyxLj
same thread: 2058626442251133383 2058659638770606211 2058663449459413237 2059480828959432755 2059482277013553613 2059482463559451015 2059485255191728629 2059486725228069323
cited on: claude-opus-4-7
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.