What is Waymo's crash rate per million miles?
Waymo's driverless fleet in California filed 1,133 NHTSA crash reports across 133.3 million driverless miles reported to the California Public Utilities Commission between September 2024 and June 2026: 8.50 reports per million miles, of which 0.80 per million allege an injury. This is a rate of filings, not of fault: an operator must report any crash it learns of, including being struck while stopped. For every other operator no public mileage exists, so no comparable rate can be computed; raw filing counts measure who drives the most.
The two series
- Driverless miles reported to the CPUC133.3 million
- NHTSA reports in the same months and state1,133 reports
- Of them alleging an injury107 reports
- Reports per million miles8.50
- Injury reports per million miles0.80
- Months with both seriesSeptember 2024 to June 2026
Last twelve months on file
- July 20257.55 per million (37 reports in 4.90 million mi)
- August 20256.42 per million (36 reports in 5.61 million mi)
- September 20255.91 per million (39 reports in 6.60 million mi)
- October 20256.69 per million (53 reports in 7.93 million mi)
- November 20257.82 per million (66 reports in 8.44 million mi)
- December 20256.63 per million (56 reports in 8.44 million mi)
- January 20266.91 per million (63 reports in 9.11 million mi)
- February 20265.72 per million (47 reports in 8.21 million mi)
- March 20267.23 per million (66 reports in 9.13 million mi)
- April 20266.13 per million (58 reports in 9.47 million mi)
- May 20266.53 per million (64 reports in 9.81 million mi)
- June 20266.30 per million (61 reports in 9.68 million mi)
How this was counted
Numerator: the operator's NHTSA Standing General Order reports for driverless (ADS) operation in California, latest version of each report, by month of incident. Denominator: driverless vehicle miles the operator reports quarterly to the California Public Utilities Commission, deployment and pilot programmes together, by month. Both series are monthly and Californian, so one can be divided by the other.
The latest month in the NHTSA file is excluded from the headline rate because reports for it are still arriving; it is shown in the monthly table marked as still being filed. Months in which the operator reported no California mileage are absent, not zero.
CPUC mileage covers passenger service in California; NHTSA reports marked CA can include test vehicles. The rate is therefore an upper bound on the passenger-service filing rate, and it is a rate of reports, not of injuries caused or of blame.
Waymo is the only operator with a public monthly mileage series. Tesla, Zoox and the others publish no mileage, so no rate exists for them and none is shown.
Data as of 2026-11-07. Numbers on this page are computed from the register's database on each visit and are CC BY 4.0; cite this page. Rules on the methodology page, exports on the data page.
Where the numbers come from
The pages below hold the rows behind every figure above, with filters, downloads and the full narratives.
What people ask about this number
- What is Waymo's crash rate per million miles?
- Waymo's driverless fleet in California filed 1,133 NHTSA crash reports across 133.3 million driverless miles reported to the California Public Utilities Commission between September 2024 and June 2026: 8.50 reports per million miles, of which 0.80 per million allege an injury. This is a rate of filings, not of fault: an operator must report any crash it learns of, including being struck while stopped. For every other operator no public mileage exists, so no comparable rate can be computed; raw filing counts measure who drives the most.
- Why is there no rate for Tesla, Zoox or Cruise?
- Because no public mileage exists for them. Only operators holding a California Public Utilities Commission permit file quarterly mileage, and of those only Waymo has a series that overlaps the NHTSA reports. Dividing filings by a company's own press-release mileage would not be reproducible.
- Is a lower rate safer?
- Not on its own. A filing is any crash the operator learned of, including crashes caused entirely by the other party, and reporting practice differs between operators. The injury rate is the more informative of the two numbers, and even that counts the operator's own severity assessment.
- Can I reuse these numbers?
- Yes. The counts are CC BY 4.0: cite The Machine Record with a link to this page. The underlying rows can be downloaded as CSV from the SGO table and the register from the data page.
Other questions with an answer
New cases, once a week, by email
Every case added to the register in the last seven days, with its sources. Free, no account, and the register is the only sender.