That’s not accurate. The original Tesla AP safety study was done on the Tesla fleet before vs. after AP was deployed. It’s about as close to a clean, controlled experiment as you can get. Same cars, same drivers, same safety equipment, roughly the same road conditions. They also didn’t even take into account whether AP was driving, only whether it was available on the car, so they aren’t looking at just the safest miles driven (highway miles).
"As a consequence, the overall 40 per-cent reduction in the crash rates reported by NHTSA following the installation of Autosteer is an artifact of the Agency’s treatment of mileage information that is actually missing in the underlying dataset."
Of course, the original publication of NHTSA's botched study is still the number one HN search result for both "NHTSA" and "Autopilot," and it still routinely gets referenced in comments. As the saying goes, the lie made it halfway around the world before the truth had put its pants on.
