AIBRIEF
An Insurance Institute for Highway Safety analysis of federal crash data found 1.28 police reportable Waymo crashes per million miles versus 4.06 for humans, but the same dataset shows gaps in dark conditions and newer cities.
The Insurance Institute for Highway Safety (IIHS) analyzed federal crash data from 2021 through 2024 covering roughly 50 million Waymo autonomous miles and found 1.28 police-reportable crashes per million miles against 4.06 for human drivers, a 68% lower rate. The gap held in Phoenix, San Francisco, and Los Angeles, Waymo's most established markets.
Waymo's sensor stack combines LiDAR (laser-based distance sensing), radar, and cameras, feeding software that does not get tired or distracted. The data confirms the stack works in the conditions it was built for: mapped, daylight-heavy, mature operations.
The data also shows where the mechanism has not held. Waymo vehicles had a higher share of crashes in dark conditions than human drivers, even though the sensor stack does not depend on visible light. In Austin, where Waymo was still in limited pre-public testing, the per-mile crash rate was slightly higher than the human baseline.
IIHS excluded minor fender benders because autonomous operators must report low-severity incidents while human drivers typically do not. Earlier Waymo-cited safety gains came from peer-reviewed but company-originated studies; the IIHS figure is the first independent benchmark at this scale.
Waymo runs robotaxi service in 11 U.S. cities; the next AV safety number will be read against the conditions this study defined.