Waymo’s driverless robotaxis had 68% fewer crashes per mile than human drivers in four US cities, a study by the Insurance Institute for Highway Safety (IIHS) has found. The IIHS also concluded that the national crash reporting system for automated vehicles needs significant improvement.

Crash rates per mile were 76% lower in Phoenix, 71% in Los Angeles and 35% in San Francisco than human drivers. Austin was an exception at 4% higher, attributed to small sample sizes. Waymo vehicles were involved in 85% fewer single-vehicle crashes and 81% fewer injury crashes per mile.

Waymo’s Level 4 driverless robotaxi service covers Phoenix, San Francisco, Los Angeles and Austin. The study period covered approximately 50m Waymo driverless miles against roughly 222bn by human drivers in the same cities.

IIHS President David Harkey said that on a limited scale, driverless cars were demonstrably safer than human drivers, but that the national data collection system was insufficient for monitoring a large-scale expansion. Of 736 automation-engaged crashes reported to the National Highway Traffic Safety Administration (NHTSA) between 2021 and 2024, only 22% were police-reportable.

The discrepancy reflects stricter reporting requirements for automated vehicle operators compared with private drivers. Eric Teoh, IIHS director of statistical services, said better data collection was essential to ensure the safety record continues as the technology scales.

Why this matters:

• 68% fewer crashes than human drivers is the most credible safety comparison yet published.Previous Waymo safety claims relied on internal data, but this is an IIHS study with independent methodology, accounting for reporting threshold differences and eliminating non-comparable incidents. Thus, it carries a weight of institutional credibility that self-reported figures cannot.
• The Austin anomaly deserves scrutiny.A 4% higher crash rate in Austin against the trend in Phoenix, San Francisco, and Los Angeles likely reflects small sample size, but it also suggests performance varies by city in ways that matter for how safety claims should be qualified as deployment expands to new operating environments.
• The data infrastructure problem may be the most consequential finding.IIHS identifying that the only national crash monitoring system for AVs cannot reliably support continuous safety oversight—because mileage reporting is voluntary and crash thresholds are inconsistent—means the regulatory foundation for scaling autonomous deployment is not yet fit for the task, regardless of how the safety numbers look today.