Self-driving cars have spent years promising to be safer than human drivers. Now there's real data to back that up — and it comes with some serious caveats.
On July 23, 2026, the Insurance Institute for Highway Safety (IIHS) released a comprehensive study comparing Waymo's driverless vehicles against human-driven cars across four U.S. cities. The result: Waymo's robotaxis were involved in crashes 68% less often than human drivers covering the same roads. It's the strongest independent safety endorsement Waymo has received to date, and it lands at a moment when regulators, insurers, and everyday drivers are all trying to figure out how much to trust autonomous vehicles.
But the same study also found a problem with the data itself: the national system used to track AV crashes isn't built to handle the technology at scale. That combination — good news about safety, bad news about oversight — is what makes this report worth paying attention to.
What the IIHS Study Actually Found
The IIHS analyzed roughly 50 million miles driven by Waymo's Level 4 autonomous vehicles between 2021 and 2024 in four cities: San Francisco, Phoenix, Los Angeles, and Austin. Researchers compared Waymo's crash involvement against human-driven vehicles operating in those same cities during the same years, using a dataset of approximately 222 billion human-driven miles for comparison, as first detailed by Axios.
The headline number is striking, but the category-by-category breakdown is even more telling:
- Overall crash involvement: 68% lower than human drivers
- Single-vehicle crashes: 85% lower
- Injury-causing crashes: 81% lower
- Waymo striking another vehicle from behind: 91% lower
- Waymo being rear-ended by another driver: 40% lower
Geography mattered too. Waymo's crash rate was 76% lower than human drivers in Phoenix, 71% lower in Los Angeles, and 35% lower in San Francisco. Austin was the outlier, showing a 4% higher rate for Waymo — a result IIHS attributed to a much smaller sample size in that market rather than a genuine safety gap.
Why the Data Isn't as Clean as It Looks
Here's the part that didn't make as many headlines: IIHS also warned that the current system for tracking autonomous vehicle crashes has real gaps. Of 736 automation-engaged crashes reported to the National Highway Traffic Safety Administration (NHTSA) between 2021 and 2024, only 22% were police-reportable incidents. The rest relied on self-reporting and inconsistent documentation standards.
That matters because as robotaxi fleets expand into more cities, insurers, regulators, and safety researchers all need reliable, apples-to-apples crash data to do their jobs. Right now, that infrastructure is playing catch-up with the technology. Industry group NAMIC (National Association of Mutual Insurance Companies) echoed this concern, noting that insurers still need more credible, standardized claims data before they can price AV risk with confidence.
Level 4 Autonomy: A Quick Refresher
Waymo's vehicles operate at SAE Level 4 automation, meaning the car handles all driving tasks within a defined operational area without a human safety driver behind the wheel. This is different from the Level 2 driver-assist systems found in most consumer cars today — think Tesla Autopilot, GM Super Cruise, or Ford BlueCruise — which still require an attentive human ready to take over. Level 4 systems like Waymo's only operate within geofenced areas (major cities, for now), which is part of why this study's city-by-city breakdown is so relevant.
The Timing Complicates the Story
Days after the IIHS study was published, a Waymo vehicle was involved in a crash that drew fresh scrutiny to the company's safety record — a reminder that even strong statistical trends don't mean autonomous vehicles are crash-proof. Waymo has also recently expanded testing to new markets, including Philadelphia, putting its safety data under an even wider spotlight as it grows.
What This Means for Everyday Drivers and Riders
If you've been hesitant to hop in a Waymo robotaxi, this study is a meaningful data point in the "for" column — but it's not the final word. A few things are worth keeping in mind:
- The data covers specific cities and conditions. Results in Phoenix's wide, sun-belt roads don't automatically translate to snow, dense urban chaos, or unmapped rural highways.
- Sample size still matters. Austin's higher crash rate is a good example of how a small number of incidents can swing a percentage dramatically.
- Reporting gaps mean some caution is warranted. With only 22% of automation-engaged crashes being police-reportable, there's more happening than official statistics fully capture — in either direction.
- This is Waymo's data, not the whole industry's. Other AV operators — Cruise (before its restructuring), Zoox, and various robotaxi startups — haven't been studied at this scale, so this shouldn't be read as a blanket statement about autonomous vehicles generally.
Why This Study Matters for the Industry
For automakers and tech companies racing to deploy autonomous systems, this study is ammunition for the argument that Level 4 autonomy, done carefully and in the right operating conditions, can outperform human driving on safety. That's a significant claim, and it arrives as regulators in multiple states debate how quickly to expand robotaxi operating permits.
At the same time, the data quality warning is a signal to regulators: safety claims from any AV operator — including Waymo — need better independent verification infrastructure before large-scale expansion should be taken purely on faith. Expect this tension to shape policy debates around autonomous vehicle deployment through the rest of 2026.
Frequently Asked Questions
Is Waymo actually safer than human drivers?
According to this IIHS study, Waymo's driverless vehicles were involved in significantly fewer crashes than human drivers across the same routes and time periods — 68% fewer overall, with even larger reductions in injury crashes. However, the study's authors caution that current AV crash-reporting systems have gaps, so this should be read as strong evidence rather than a definitive final verdict.
What cities did the IIHS study cover?
The study analyzed Waymo operations in San Francisco, Phoenix, Los Angeles, and Austin between 2021 and 2024.
What is SAE Level 4 autonomous driving?
Level 4 means the vehicle can handle all driving tasks without human intervention, but only within a specific, geofenced operating area — unlike Level 5, which would work anywhere, or Level 2 systems that still require an attentive driver.
Why was Austin's data different from other cities?
Waymo showed a 4% higher crash rate in Austin compared to human drivers, which IIHS attributed to a smaller sample size in that market rather than an actual safety shortfall.
Does this study mean all self-driving cars are safer than humans?
No. The study specifically examined Waymo's Level 4 vehicles operating in four U.S. cities. Other autonomous vehicle operators, including Zoox and various robotaxi startups, have not been analyzed at this scale, so the results shouldn't be generalized across the entire AV industry. Each company's technology, operating conditions, and safety record need to be evaluated independently.
What are the data quality concerns raised by IIHS?
Of 736 automation-engaged crashes reported to NHTSA between 2021 and 2024, only 22% were police-reportable, meaning much of the data relies on self-reporting rather than independently verified records. IIHS says the national tracking system isn't yet built to handle AV safety monitoring at scale.
Conclusion
The IIHS study gives Waymo its most credible safety endorsement yet, backed by 50 million real-world miles of data. Crash rates 68% lower than human drivers — with even steeper drops in injury crashes — are hard to dismiss. But the same report is a reminder that the infrastructure for verifying AV safety claims is still catching up to the technology itself. As robotaxis expand into more American cities, expect both halves of this story — the safety data and the data-quality gaps — to keep shaping the conversation around autonomous vehicles.
Suggested Internal Links
- Link to any existing ADAS/Level 2 driver-assist explainer articles
- Link to prior Waymo or robotaxi coverage, if published
- Link to EV/technology category hub
- Link to any Tesla Autopilot or Full Self-Driving coverage for contrast
Sources to Verify
- IIHS official study: https://www.iihs.org/news/detail/waymos-driverless-cars-crash-less-often-than-people
- Axios: https://www.axios.com/2026/07/23/waymo-robotaxis-av-crash-study
- Electrek (caveats analysis): https://electrek.co/2026/07/25/waymo-is-2-3-safer-than-a-human-driver-says-iihs-with-some-caveats/
- Torque News (NAMIC data quality response): https://www.torquenews.com/17998/iihs-says-waymo-data-point-lower-crash-rates-limited-deployments-namic-says-insurers-still
- The Philadelphia Inquirer (Waymo expansion context): https://www.inquirer.com/transportation/waymo-human-driver-safety-study-philadelphia-20260728.html
Social Media Excerpt (under 280 characters)
New IIHS study: Waymo's driverless cars crash 68% less than human drivers across 50M miles. But the report also flags major gaps in how the U.S. tracks AV safety data. The full breakdown 👇 #Waymo #SelfDrivingCars #AutonomousVehicles
