Don't know that it's a fallacy, it's more that the question you ask is impossible to answer - it's impossible to have a number for how many crashes were prevented. But we do have a number for crashes caused by. So we work off of that - if Telsa want's to argue they are safer, I'd love for them to devise some way to prove it, but I can't see that happening.
It seems like comparing AP to anything but regular cruise control miles is not very fair and is going to inflate its safety.
From that link, it's pretty bad if it's not even clearly better at general "highway miles" than when it's off because people are least likely to use it at the times that are much more dangerous generally, such as merging onto and off the freeway.
