IIHS Study Finds Waymo Autonomous Taxis Crash 68% Less Than Human Drivers, But Notes Key Caveats

According to the Insurance Institute for Highway Safety (IIHS), autonomous electric taxis operated by Waymo crash 68% less frequently than the average human driver. The safety research organization, widely recognized for its "Top Safety Pick" crash recommendations, examined federal crash reporting data spanning 2021 through 2024 to evaluate the performance of Level 4 autonomous vehicle services.

Level 4 automation allows vehicles to operate without a driver present within defined geofenced areas. Waymo, a subsidiary of Google's parent company Alphabet, Inc., currently offers autonomous ride-hailing services across 11 U.S. cities, covering an aggregate geographical area larger than the state of Rhode Island. Having completed tens of millions of trips, Waymo provided a substantial real-world dataset for evaluation.

The IIHS study focused primarily on Waymo due to data availability from other operators during the 2021–2024 evaluation window. Rival operator Cruise halted operations in 2023, while Zoox only initiated public rides in 2025. Tesla was similarly excluded from the core study window because its Austin service only launched in 2025 with safety drivers inside, accumulating roughly 380,000 total unsupervised miles to date.

Comparing the Crash Rates

After filtering the raw federal data to eliminate redundancies and isolate incidents severe enough that an average person would report them to law enforcement, the IIHS evaluated a sample size of 50 million autonomous miles driven by Waymo vehicles.

The data revealed that human drivers logged an average of 4.06 police-reportable crashes per million vehicle miles traveled. In contrast, Waymo recorded 1.28 crashes per million miles traveled. The crashes involving Waymo vehicles tended to be less severe on average and were generally determined not to be the fault of the autonomous vehicle.

Geographically, Waymo outperformed human safety averages across three of the four analyzed markets: Phoenix, San Francisco, and Los Angeles. However, in Austin, Waymo logged a slightly higher crash rate than human drivers. The IIHS attributed this outlier to a relatively small sample size of miles driven during the testing phase prior to public availability in 2024.

Additionally, despite relying on sensor suites including radar and LiDAR—technologies that operate independently of visible light—Waymo recorded a higher proportion of its crashes in dark conditions compared to human drivers.

Research Context and Technical Caveats

The independent IIHS findings mirror prior research published by Waymo. A 2023 study authored by Waymo employees reported an 85% reduction in injury-causing crashes and a 57% reduction in police-reported crashes. A subsequent 2025 study claimed Waymo vehicles were 25 times safer for pedestrians and cyclists. While those earlier studies were submitted for peer review, the IIHS investigation offers independent third-party confirmation.

However, the IIHS highlighted several important caveats that contextualize the headline numbers:

• Reporting Distortions: While autonomous vehicle companies are mandated to report all crashes, they are not legally required to report autonomous miles driven. Waymo voluntarily discloses mileage data, enabling crash rate calculations. Because human drivers frequently do not report minor scrapes or fender benders to police, IIHS researchers had to manually clean the dataset to ensure fair comparisons.
• The "Average Human" Baseline: The study compares Waymo to average human drivers, a group that includes distracted, overworked, or impaired individuals. The IIHS noted that comparing peak-capacity automated systems against optimal human driving standards would provide a clearer picture of potential vs. actual safety performance.
• Operating Environments: Waymo does not operate on high-speed highways. Because highways feature simpler traffic flows and lower overall crash rates per mile than urban streets, highway miles were excluded for both humans and Waymo. Furthermore, 50% of Waymo's crashes occurred on streets with speed limits under 25 mph, compared to just 8% for human drivers, directly contributing to lower crash severities.
• Occupancy Factors: Nearly half of Waymo’s logged miles were driven without passengers inside the vehicle. Without human occupants, internal vehicle injury counts naturally drop. However, the IIHS noted that unoccupied vehicles driving extra miles to reposition could increase general traffic congestion.
• Sample Size Disparities: While 50 million miles represents a large dataset for autonomous testing, it remains small compared to human driving volume. Human drivers log over 3 trillion miles annually in the U.S. and drove approximately 222 billion miles within the study areas during the same timeframe. Due to Waymo's smaller sample size, specific rare crash categories contained only zero or one recorded incident, creating localized statistical spikes.

While the IIHS emphasizes that broader datasets are needed to draw definitive long-term conclusions, the initial 50-million-mile dataset indicates that autonomous taxis are currently demonstrating significantly lower crash frequencies than average human drivers in urban environments.