Driverless car safety is becoming one of the most important transportation questions of 2026. For years, autonomous vehicles were discussed mostly as a future technology. Now, fully driverless robotaxis are operating on public streets in several U.S. cities, which means researchers finally have enough real-world mileage to begin comparing their crash performance with human drivers.
A new Insurance Institute for Highway Safety study offers one of the clearest comparisons so far. Researchers found that Waymo vehicles operating without a human driver had a police-reportable crash involvement rate that was 68% lower than the rate for human drivers in the same cities and time periods. The study also found large reductions in injury crashes and single-vehicle crashes.
Those results are encouraging, but they do not prove that every autonomous vehicle is safer than every human driver. The study focused mainly on Waymo’s Level 4 driverless operation, and researchers emphasized that better national data collection is still needed. For readers following broader crash trends, our Accident Statistics hub provides context on how safety rates should be interpreted, while our Major Incidents & Safety Analysis section covers emerging transportation risks.
What the 2026 Driverless Car Safety Data Shows
Waymo’s Driverless Crash Rate Was Lower Than the Human Driver Rate
The IIHS study examined crashes involving highly automated vehicles and compared them with police-reported crashes involving human drivers. To make the comparison more realistic, researchers focused on incidents that a typical human driver would likely have reported to police instead of counting every minor contact recorded by an autonomous fleet.
That adjustment matters because driverless fleets operate under different reporting requirements. Companies may have to document incidents that a private driver would never report, including minor property damage or low-speed contacts. A simple raw crash count can therefore make automated vehicles appear worse than they are if the reporting standards are not aligned.
Waymo Had 68% Fewer Police-Reportable Crashes per Mile
After cleaning the data and matching comparable types of crashes, IIHS researchers found that Waymo’s driverless vehicles had a crash involvement rate 68% lower than human drivers across the areas and years studied. The comparison included operations in Phoenix, San Francisco, Los Angeles, and Austin.
The researchers used roughly 50 million miles of Waymo driverless operation and compared that exposure with about 222 billion miles driven by humans in the same locations and time periods. Looking at rates per vehicle mile traveled is essential because a company that drives more miles will naturally have more opportunities to be involved in crashes.
The results also varied by city. Waymo’s police-reportable crash rate was substantially lower than the human rate in Phoenix and Los Angeles and lower in San Francisco. Austin showed a slightly higher rate for Waymo, but the sample there was much smaller. That variation is a reminder that autonomous performance can depend on location, road design, operating conditions, and the amount of data available.
Injury and Single-Vehicle Crash Rates Were Also Lower
The study found that Waymo’s driverless vehicles were involved in 81% fewer injury crashes per mile than human drivers. They were also involved in 85% fewer single-vehicle crashes. In another notable finding, their rate of rear-ending another vehicle was 91% lower.
These patterns make sense when considering some of the risks automation can remove. A robotaxi does not become drunk, tired, distracted by a phone, angry in traffic, or impatient after a long shift. Automated systems can continuously monitor multiple directions and do not experience the same attention lapses that affect people.
However, automation introduces different risks. Sensors can be affected by unusual conditions. Software must interpret complicated road behavior. Construction zones, emergency scenes, unexpected human actions, and rare edge cases can challenge automated systems in ways that are different from ordinary human error.
Why Comparing Driverless Vehicles With Humans Is Harder Than It Sounds
One of the most important parts of the 2026 research is not the 68% figure itself. It is the work researchers had to do to create a fair comparison. Crash reporting for autonomous fleets and human drivers is still inconsistent, which makes simple comparisons misleading.
Reporting Rules and Mileage Data Can Distort the Numbers
Federal reporting requirements require companies covered by the rules to report certain crashes involving automated driving systems. Human drivers, by contrast, do not report every scrape, bump, or low-damage collision. Many crashes that technically qualify for reporting may never appear in police data.
IIHS researchers reviewed hundreds of automated-vehicle crash reports and removed duplicate records, incidents where the automation was not engaged, events that did not occur on public roads, and cases that did not represent a meaningful crash. They then estimated whether a reasonable person would likely have reported the event to police.
Only about 22% of the 736 public-road crashes reviewed were considered likely police-reportable or possibly police-reportable. That difference shows why counting all reported autonomous-vehicle incidents and directly comparing them with police crash totals for human drivers would not produce a reliable safety conclusion.
Mileage is another limitation. Waymo voluntarily provides enough driverless vehicle-mile data to calculate a crash rate, but comparable exposure data is not consistently available for every automated-driving company. Without knowing how many miles a fleet travels, researchers cannot fairly determine how often it crashes.
What Driverless Car Safety Data Means for Drivers and Cities
Driverless Technology Is Not the Same as Driver-Assist Technology
Consumers should be especially careful about one common source of confusion: a fully driverless Level 4 robotaxi is not the same thing as the driver-assistance system available in many privately owned cars. Mixing these technologies together can create dangerous expectations.
Level 4 Robotaxis and Level 2 Driver Assistance Require Different Behavior
Waymo’s vehicles in the study operate at SAE Level 4 within defined operating conditions. In driverless service, the automated system performs the driving task without a human sitting behind the wheel and continuously supervising it. That is fundamentally different from Level 2 systems that combine functions such as adaptive cruise control and lane centering.
With Level 2 driver assistance, the human driver is still responsible for supervising the road and must remain ready to intervene. Treating a Level 2 system as if it were a self-driving vehicle can create a serious safety problem because the system may encounter conditions it cannot manage on its own.
That distinction should remain central as automated technology expands. A strong safety record for one Level 4 fleet does not automatically prove that every automated system, driver-assistance feature, or future autonomous vehicle will perform equally well. Software, sensors, operating areas, fleet procedures, and safety culture can all differ.
Cities also need better data as driverless fleets expand. Researchers need consistent mileage information, comparable crash definitions, and enough detail to evaluate where and why crashes happen. The more autonomous vehicles operate in new weather, traffic, and road environments, the more important continuous safety monitoring becomes.
everyday driver
For everyday drivers, the practical lesson is straightforward: do not assume that every vehicle marketed with advanced technology is capable of driving itself. Read the manufacturer’s instructions, understand what the system can and cannot do, and remain responsible whenever supervision is required. Our Safety Resources page includes additional guidance for reducing everyday road risk, and our article on U.S. pedestrian deaths in 2025 explains why safer vehicles must still operate within safer streets.
The 2026 IIHS findings are an important milestone. They suggest that Waymo’s current Level 4 driverless vehicles, in the locations studied, are involved in substantially fewer police-reportable crashes per mile than human drivers. That is meaningful evidence that well-designed automation can potentially reduce some of the risks created by human error.
At the same time, the correct conclusion is not that the driverless-car safety debate is over. The data applies to a specific fleet, operating in specific places, during a specific period. As autonomous vehicles expand into more cities and more complex environments, transparent reporting and independent evaluation will be essential.
Readers who want to review the underlying research can consult the Insurance Institute for Highway Safety study on driverless vehicle crash rates. The most useful takeaway is cautious but significant: some Level 4 driverless vehicles are now showing a lower real-world crash involvement rate than human drivers, but continued measurement will determine whether that advantage holds as the technology scales.