Waymo says its autonomous-driving system was involved in substantially fewer injury-related crashes than human drivers in the same operating areas and period, according to a company safety analysis covering more than 220 million fully autonomous miles through the end of March 2026.

The comparison covers five operating geographies and includes Atlanta for the first time. Waymo reports 94% fewer crashes causing serious or fatal injuries, 82% fewer crashes involving airbag deployment and 82% fewer crashes involving any reported injury than the human-driver comparison group.

These are company-reported crash-involvement rates, not an independent finding that Waymo caused fewer collisions. The comparisons include crashes regardless of fault.

Contents

What the analysis found

The analysis compares Waymo Driver operations with human-driver crash rates in the same areas and over the same period. This type of exposure-normalised comparison is important because raw crash totals can be misleading: a group that drives more miles will generally have more opportunities to be involved in a crash.

Across the overall dataset, Waymo reports:

  • 94% fewer crashes causing serious or fatal injuries;
  • 82% fewer crashes involving airbag deployment; and
  • 82% fewer crashes involving any reported injury.

The company also reports lower rates for crashes involving vulnerable road users. It says injury-causing crashes involving pedestrians were 93% lower, while those involving cyclists and motorcyclists were each 84% lower than the human benchmark.

The latest analysis differs from earlier updates in both scale and operating conditions. Waymo says the service area now includes a unified San Francisco Bay Area operation, expansion into more complex environments such as airport service, and the addition of Atlanta.

In Atlanta, the Waymo Driver accumulated more than 5.4 million fully autonomous miles. Compared with the human benchmark, Waymo reports 94% fewer crashes involving airbag deployment and 86% fewer crashes involving any injury. The company says both Atlanta comparisons were statistically significant.

Waymo also gives a less common-event example from Atlanta: it estimates that a typical human driver covering the same 5.4 million miles would have been expected to experience approximately 1.2 crashes causing serious or fatal injuries. The Waymo Driver experienced none of those serious-or-fatal-injury crashes in the dataset.

The company says it now drives more than 4 million miles each week. Based on its comparisons, Waymo estimates that its operations correspond to one fewer serious-injury-or-worse crash every eight days, about six fewer airbag-deployment crashes each week and about 13 fewer crashes involving any injury each week.

Over the lifetime of its operations, Waymo estimates that its autonomous miles correspond to 47 fewer crashes causing serious or fatal injuries, 305 fewer crashes involving airbag deployment and 707 fewer crashes involving injuries of any kind than would have been expected if the same miles had been driven by humans in the same places.

Why exposure and location matter

The central question in this analysis is not simply how many crashes occurred. It is how often crashes occurred relative to the distance travelled, and whether the comparison is made against human driving in comparable places and periods.

A crash-involvement rate can include a collision in which the autonomous vehicle was not at fault. That distinction matters because the reported percentages describe how often Waymo vehicles were involved in specified types of crashes, rather than proving that the system caused or prevented every event included in the calculation.

The comparison uses human-driver crash rates from the same areas and period. However, the supplied announcement does not describe how differences in road environment, road type, traffic conditions, vehicle use, reporting practices or other exposure characteristics were handled. The same operating area does not automatically mean that every exposure is equivalent.

The scale of the dataset is nevertheless important. More than 220 million fully autonomous miles represent a large and continually expanding operational exposure, while the addition of Atlanta gives the update a new location-specific comparison alongside the aggregate results.

What the figures do not establish

The evidence supplied for the analysis is a Waymo announcement published on June 24, 2026.

The analysis is observational, not a randomised experiment. The source does not specify the complete population of human drivers, the datasets used to calculate human crash rates or the precise matching procedure. It also does not provide the underlying crash counts for every percentage, confidence intervals, p-values, the statistical model or the significance threshold.

That information is particularly important for rare outcomes. A small number of serious crashes can produce a large percentage difference, and the stability of such an estimate depends on the number of events and the uncertainty around the comparison. Waymo's statement that the two Atlanta results were statistically significant indicates that they passed a specified statistical test, but the supplied material does not provide enough detail to assess that test independently.

The results also do not demonstrate that Waymo's performance is the same in every possible road, traffic or operating condition. They apply to the analysed fully autonomous operations in the reported geographies and period. The Waymo Driver is a service-operated autonomous system, not a consumer driver-assistance feature generally available for purchase.

Waymo says the Driver undergoes a formal safety review before deployment on public roads and refers to acceptance criteria in a separate paper. The supplied material does not identify a government approval, regulator finding or specific legal safety standard.

What to watch next

The next important step is greater methodological detail. The full data and methodology linked by Waymo would need to clarify how human crash rates were constructed, how exposure was matched, which crash definitions were used and how uncertainty was estimated.

Future updates will also show whether the reported pattern persists as Waymo adds miles, expands to new locations and operates in more complex environments. Independent analyses using comparable definitions would provide a further check on the company's conclusions.

For now, the analysis supports a narrower statement: Waymo reports lower crash-involvement rates than its human-driver benchmarks across a large exposure base and several specified outcome categories. It does not amount to a guarantee that an individual ride will be crash-free, nor does it establish identical performance outside the operating conditions examined.

Sources