from the article: "...with zero confirmed fatalities and a wealth of internal data demonstrating safer, more predictable vehicle control performance" and "That

from the article: "...with zero confirmed fatalities and a wealth of internal data demonstrating safer, more predictable vehicle control performance" and "That contrasted against worldwide accident data, customers using Autopilot are statistically safer than those not using it at all." (emphasis mine) i also stumbled over this statement, but i think they are using all telemetry data (e.g. what would the autopilot have done vs what the human actually did) in all accidents, not just fatalities per miles driven.