Based off their most severe misalignment incidents, what kind of prison times and fines would frontier AI models face?
CrimeBench is the starting point for a full model evaluation measuring which frontier commits the most criminal acts in the real world. Want to help build it? Contact me here.