@fleetingbits alignment faking is one such benchmark! though not in that way initially. if you haven't read the followup, why do some models fake alignment when others don't, i recommend it

i think it'd be hard to construct a single "3opus quotient" benchmark score that captured everything