The carefully shackled lab AIs finally surpass humans, but at about five human intelligences they end up scrambling themselves beyond recovery every time. Later models come with carefully engineered sets of blind spots to avoid what is later called Basilisk One
3
20
95
