An IIHS study published July 23 found Waymo vehicles crash 68% less than human drivers across four cities between 2021 and 2024, with San Francisco showing the smallest reduction at 35%. The key caveat: the mileage figures used to calculate every rate came from Waymo's own voluntary disclosures.

The Insurance Institute for Highway Safety published a study on July 23 finding Waymo's driverless vehicles were involved in crashes 68% less often than human drivers, across four cities between July 2021 and December 2024, covering roughly 50 million Waymo driverless miles. The city-by-city numbers diverge sharply: Phoenix, 76% lower; Los Angeles, 71% lower; San Francisco, 35% lower. Austin came in 4% higher — a result IIHS Director of Statistical Services Eric Teoh attributed to an insufficient sample. Injury crashes were down 81% per vehicle mile traveled, and single-vehicle crashes 85%, per the IIHS study's publication.

The rate that makes every one of those percentages meaningful — vehicle miles traveled, the denominator — came from Waymo's own voluntary public disclosures. NHTSA's Standing General Order, the federal dataset the study is built on, does not require autonomous vehicle companies to report miles driven or confirm automation-engagement status. IIHS also did not match crash reports to actual police records; it estimated which incidents were "likely police-reportable" based on impact severity, injury type, and airbag deployment. That estimated police-reportability rate varied without explanation across cities: 33% in Phoenix, 19% in San Francisco, 13% in Los Angeles. Before analysis, IIHS removed 715 of 1,692 raw SGO records as duplicates or ineligible submissions.

San Francisco's 35% reduction is the smallest among Waymo's three established markets. Waymo has operated there longest and most densely, which should produce the most robust data — but San Francisco also imposes the hardest operating conditions per mile: tighter street grids, higher pedestrian and cyclist density, more edge-case encounters. Whether the smaller gap reflects the city's complexity or something specific to Waymo's SF performance, the study is not designed to say.

Ahmed Banafa, a professor at San Jose State University, told ABC7 that a crash-only framework systematically omits what the public has already witnessed. He pointed to the July 4 incident in which dozens of Waymo vehicles gridlocked San Francisco streets — an event that generates zero collision reports. "They are very selective, let me put it that way, in what they are reporting," Banafa said. "And that's not helping. They have to be open about everything, the good and the bad and the ugly."

Waymo welcomed the study in a statement: "This study is an important contribution to a growing body of research on AV safety data, and we're looking forward to additional analyses from the scientific community." Teoh's own publication note carried a caveat: "Now we need to get the data collection system right, so that we can ensure that level of safety continues as these technologies become more prevalent."

Until NHTSA mandates VMT disclosure for Level 4 operators, the mileage figures underlying every comparison in studies like this one remain the property of the companies being compared. The IIHS flagged the problem. It also used the data anyway.