IIHS Says Waymo’s Driverless Cars Appear Safer Than Humans—If You Squint at the Numbers

When the Insurance Institute for Highway Safety (IIHS) announced that Waymo’s driverless fleet crashes less often than a human behind the wheel, some people cheered while others raised an eyebrow. After all, saying a robotaxi is "safer" sounds great—until you dig into the numbers and realize you’re squinting through a lot of statistical smoke.

First, the data problem. Getting reliable crash stats for human drivers is…well, easy. Every police report, every insurance claim—there’s a mountain of it dating back decades. Autonomous‑vehicle crash reports, on the other hand, are a different beast. California only started requiring companies to log every incident in 2014, and the federal NHTSA didn’t demand consistent reporting until 2021. Even then, the rules for a self‑driving car aren’t identical to those for a human‑piloted one.

One metric that matters a lot is crashes per vehicle‑mile traveled. The Department of Transportation publishes annual vehicle‑miles‑traveled (VMT) numbers for the whole country, but there’s no comparable mandatory reporting for robotaxis. Waymo, being unusually transparent, posts its own mileage figures; most other firms keep theirs under wraps. That’s why the IIHS study focuses almost exclusively on Waymo and leaves out Tesla, Cruise, Zoox and the rest.

Now, onto the messy part: cleaning the data. IIHS had to wade through crash reports from 2021‑2024, toss out duplicates, strip away incidents where the automation wasn’t actually engaged, and discard anything that didn’t happen on a public road. They also applied a common‑sense rule for police‑reportable crashes—if a passenger felt a pain and drove off, that incident got coded as “unlikely to be reported,” but an airbag deployment automatically counted as police‑reportable, regardless of how minor the damage was.

After this pruning, less than a quarter of the original autonomous‑vehicle incidents survived the gauntlet. From an initial pool of 736 public‑road crashes where the automation was on, only 22 % (about 162) were deemed usable. Those broke down into 89 Waymo crashes, 50 Cruise, 10 Zoox, and 10 from other companies. Because mileage data existed only for Waymo, the institute used those 89 incidents to calculate a crash‑rate per mile.

The headline‑grabbing conclusion? Waymo’s crash involvement rate was roughly 68 % lower than that of human drivers overall—76 % lower in Phoenix, 35 % lower in San Francisco, 71 % lower in Los Angeles, but oddly 4 % higher in Austin (a tiny sample). Single‑vehicle crashes fell by 85 % and injury crashes by 81 % compared with human‑driven traffic.

Sounds impressive, right? Hold on a second. The raw numbers tell a very different story. NHTSA recorded about 24.35 million police‑reported crashes involving human drivers from 2021‑2024. Waymo, by contrast, logged just 89 police‑reportable incidents in the same period. In terms of mileage, Waymo’s driverless cars logged roughly 50 million miles, while human drivers covered an astronomical 222 billion miles in the same cities.

Statistically, comparing a sample of 89 crashes to millions of human‑driver crashes is like trying to judge a movie’s quality based on a single scene. The sample size is minuscule, and the operating environments differ—Waymo stays in dense urban cores, drives slower, and avoids highways, which naturally reduces crash exposure.

IIHS itself acknowledges the limitation, noting that “the method of coding narratives is unsustainable as deployments and incident numbers grow.” They recommend better national data collection for Level‑4 autonomous vehicles so future safety assessments can be more robust.

So, are Waymo robotaxis a public‑health breakthrough or just a statistical illusion? The answer probably sits somewhere in the middle. The data suggests they’re doing something right, but the current evidence isn’t strong enough to claim they’re definitively safer than human drivers across the board.