I suppose you count those as wins for yourself, but it seems smart and safe to insist on turning over control in that situation. Humans don't have a safe failure mode like that where they will stop driving or surrender control to an automated system when intoxicated, falling asleep, and so on.
I'm arguing that you can't point to numbers and say "Look, AP is safer than humans, because in the optimal subset of conditions where AP is 'able' to drive, it does better than humans in all conditions, even ones where AP would refuse to engage".
