Summary: Waymo's latest safety data shows dramatic crash reductions across 170.7 million rider-only miles, but the methodology behind those headline percentages deserves closer scrutiny before we declare self-driving cars universally safer than humans.
Waymo has driven 170.7 million rider-only miles without a human behind the wheel through December 2025. That number alone sounds impressive. But the real attention-grabber is the claim attached to it: 83% fewer airbag deployment crashes compared to the average human driver over the same distance. So what does that actually mean, and just as importantly, what does it leave out?
What Rider-Only Miles on Surface Streets Really Covers
Let's start with the comparison basis, because this matters more than most people realize. Waymo's crash reductions are measured against human driving on surface streets only, and they count crashes regardless of who was at fault. That is a specific slice of driving. It excludes highways. It excludes conditions where Waymo simply does not operate.
The mileage itself breaks down across four cities: Phoenix leads at 68.6 million miles, followed by the San Francisco Bay Area at 53.5 million, Los Angeles at 37.9 million, and Austin at 10.7 million. Waymo frames the scale by saying that 127 million fully autonomous miles equals over 150 human driving lifetimes of experience. That is a clever way to make the number feel tangible. But a lifetime of driving in one city is not the same as a lifetime of driving across rain, snow, rural roads, and every other condition humans handle daily.
The Numbers Shift as Miles Grow
Here is something worth paying attention to. As of June, when Waymo's fleet had covered 96 million miles, the company claimed 91% fewer crashes resulting in serious injury compared to an average human driver over the same distance. By December 2025, at 170.7 million miles, that figure shifted to 92% fewer serious injury or worse crashes. The numbers are close, but they are not frozen. They evolve as the dataset grows and operating conditions change.
The broader picture includes 82% fewer injury-causing crashes, or 544 fewer than the human benchmark would predict. Vulnerable road user numbers stand out too: 92% fewer pedestrian crashes with injuries, 85% fewer cyclist crashes with injuries, and 81% fewer motorcycle crashes with injuries. All results have been rounded to the nearest whole number, which is standard practice but worth remembering when you see precise-sounding percentages.
Not Every Crash Makes the Headline
The aggregate percentages paint one picture. The individual crash reports paint another. Between mid-February and mid-August, Waymo vehicles were involved in 45 crashes reported to the government. Of those, 24 involved the Waymo not moving at all, and 7 involved the Waymo being rear-ended by another vehicle. Three crashes involved a Waymo passenger opening their door without looking, injuring a bicycle or scooter rider. One involved a wheel falling off a Waymo vehicle.
These details matter because the 83% figure folds all of this together. A stationary car getting rear-ended counts the same way in the comparison.
What We Still Do Not Know
The sources do not explain how Waymo calculates the human benchmark crash rate in detail. We also do not get specifics on what qualifies as a "serious injury" in Waymo's classification, or how factors like weather, time of day, and road type might differ between Waymo's operations and the human driving baseline.
None of this means the data is worthless. It means the data is incomplete as a universal safety claim. The context around these numbers is narrower than the headlines suggest.
The Bigger Picture in Autonomous Driving
Waymo is accumulating miles and publishing data in a way its competitors are not. Tesla reported 3 crashes to NHTSA in the first month of its robotaxi launch with a fleet of just 10 to 20 vehicles covering 7,000 total miles. By comparison, Waymo's dataset is massive and growing.
But the question is not whether Waymo is safer than other autonomous driving companies. The question is whether "83% fewer crashes on surface streets" translates to "83% safer than a human driver in your daily life." Those are very different claims, and the data we have right now only supports the first one.
So what would it take for you to trust a self-driving car with your family? Is a surface-street-only comparison enough, or do you need to see highway numbers, snow performance, and a clear explanation of the human baseline first?
Comments