Teslas on autopilot are involved in fewer accidents per mile traveled (adjusted, as much as possible, for type of driving) than human drivers
Not a slam on Tesla here, I totally think they are in the right on this so far every autopilot incident I am aware of the operator was misusing the technology by not paying attention and being prepared to take over as Tesla makes it VERY CLEAR every time you turn the thing on that you are supposed to be.
That said the statistic is a bit self serving because most users would be driving manually in the most challenging conditions because autopilot can't or won't. I don't know to many human drivers who get into accidents while cursing down long strait interstate highways in daytime fare weather conditions with minimal to moderate traffic. AP gets used a lot in that situation I would expect. Add some bad weather, cross streets, traffic, other aggressive drivers, poor or absent markings, construction, cops direction traffic, etc and the rate of human error I have no doubt goes up sharply. This would all be things that AP does not deal with much in terms of the "miles its driven" but human drivers do deal with.
So when Tesla says that AP gets into less accidents per mile than human drivers, and they have, I find it suspect that while it might be technically correct (the best kind of correct :-)) in the strictest sense, it might not really mean much.