"AI drivers" glitch/fail differently than human drivers, and in ways that to humans look bizarre and easily preventable. Human driver failures (like being tired, drunk, very upset, or being on your phone, distracted by passengers, etc), to humans, look understandable, but from an AI perspective they look bizarre and easily preventable.
In theory, what we should care about is which ones cause more deaths or accidents. The fact that AI accidents seem worse based on naive intuition shouldn't matter.
However, humans serve on juries so here we go... Would there be a 200 million dollar judgement against someone who tried to get a can of pop from the back seat, ran over a pedestrian and then tried to lie about e.g. whether the light was red?