New York’s public transit advocacy community has launched a new attack against robotaxis. A recently published report from Open Plans and the Sam Schwartz Transportation Program leads with an alarming claim: Waymo vehicles, they assert, “have a significantly higher rate of crashes with a serious injury or fatality than the existing New York City for-hire fleet.”
The full report largely concedes the point that robotaxis are safer than the typical human driver (an earlier point of debate) and instead focuses on the comparison with for-hire services like Uber, Lyft, and yellow taxis. Concern about the impact on jobs in this sector is the main reason that autonomous vehicles remain illegal in New York. (Waymo’s limited testing permit lapsed in March.)
Finally, a reason to check your email.
Sign up for our free newsletter today.
But a closer look at what the report actually found shows that the safety record of robotaxis is so strong that they clear even the bar of safety compared with professional drivers in the Big Apple. Unfortunately, the report skewed the numbers to push the authors’ preferred narrative.
The study’s strongest statistical analysis looks at the risk of injury-causing crashes among different vehicle types. These data are taken from Waymo’s reporting to federal regulators and crash data from the NYPD indicating injuries or fatalities. Over 40,000 such cases occur in New York City in a typical year, allowing for a statistically sound comparison against Waymo’s record, whether deployed in San Francisco or nationwide.
The report’s key finding is that Waymos are involved in far fewer injury-causing crashes than either New York City drivers or for-hire vehicles: 0.71 crashes per million miles nationally compared with 2.15 for New York City’s taxis, Ubers, and Lyfts. This is especially striking because New York City’s lower speed limits and other infrastructure substantially reduce accidents relative to the national average. Waymo has reached a safety level that exceeds even that of New York City’s professional drivers. In fact, the authors found that Uber and Lyft drivers have slightly higher crash rates than other drivers, and yellow cab drivers have even higher crash rates than Uber and Lyft drivers. This may not be surprising because Uber and Lyft’s rating system allows for a greater reputational record than yellow taxis.
When we turn to more serious accidents resulting in death—which Vision Zero intends to eliminate—New York City has a far lower auto fatality rate than the rest of the country but still sees over 200 auto deaths a year. In a comparison of fatal crashes, Waymo, all-city drivers, and for-hire vehicles are essentially tied. Yellow taxis are slightly more deadly than New York drivers in general, while Uber and Lyft are slightly less deadly. However, given the sparse data, it’s hard to tell exactly what’s going on.
Even when judged against professional drivers in New York City—a highly demanding benchmark—and even when compared using fatalities, Waymo has a superior or similar record. The report’s authors nonetheless conclude otherwise based entirely on an analysis of “Killed or Seriously Injured” (KSI).
Waymo’s entire KSI record—across 220 million driverless miles since 2020—consists of just three crashes. New York’s for-hire fleet had 16 over 1.75 billion miles in 2025. Analyzing a variable this rare doesn’t tell us much. The report, to its credit, published 95 percent confidence intervals, which show that the crash rate for New York City’s for-hire fleet is 0.005–0.015, which overlaps with a value of 0.003–0.040 for Waymo nationally and 0.004–0.108 for the Bay Area. The authors wrote that “the data cannot rule out Waymo having a comparable KSI rate.”
A deeper look at the data makes the picture even more favorable to Waymo. The serious-injury record for Waymo includes just two fatal crashes. But in neither accident was the killed passenger riding in the autonomous vehicle. In one case, a Waymo robotaxi in San Francisco was rear-ended while completely stopped and then hit other vehicles in a multi-car collision, killing one person. In the other case, a motorcyclist hit a Waymo that was signaling a turn and yielding to a pedestrian; the cyclist was then struck by another vehicle and later died.
While these cases are tragic, Waymo was not blamed for the crashes. In fact, Waymo has been judged at fault in a tiny fraction of the accidents it has been involved in —less than 5 percent, according to one analysis.
Even including fatalities Waymo wasn’t responsible for, Waymo is still statistically much safer than New York City drivers overall. It is only at parity with for-hire drivers, who are about ten times less risky than all drivers in Open Plans’ analysis on KSI. This finding should cause some confusion. If yellow taxi drivers cause more crashes with injuries than all drivers—and lead to more fatalities than all drivers—how do they show up as ten times less risky in an analysis on serious injuries?
The problem here lies in a comparison of two dissimilar sources, as Jonathan Nolan has pointed out. The all-vehicle KSI measure (3,188 crashes) comes from applying a “modified New York State Department of Motor Vehicles (NYS DMV) severity formula” to raw NYPD collision records. This is a broad severity category that includes everything from concussions, fractures, severe lacerations, internal injuries, semiconsciousness, etc.
But for for-hire vehicles, the report uses data from the Taxi and Limousine Commission, which mandates crash reporting for licensed vehicles, where the relevant category from Local Law 31 is critical injury. This is defined as a “severe injury that poses an immediate threat to life or is likely to cause long-term impairment or disability.” These two categories are not comparable.
A simple way to see the mismatch is to examine the implied ratio of serious-injury crashes to fatal crashes. For all New York City vehicles, it’s about 13 to 1 (3,188 KSI to 252 fatal). With professional drivers, it’s 1.07 to 1 (16 to 15). If you take the data literally, it suggests that taxi drivers almost never seriously injure someone without also killing them, which is both physically absurd and exactly what you would expect if the “serious injury” definition for for-hire vehicles is actually measuring closer to fatality.
Suppose we instead make a simple assumption: that for-hire vehicles result in serious injuries in the same proportion to deaths as the citywide fleet. That would imply roughly 190 KSI crashes, not 16, at a rate of 0.11 per million miles. By that measure, Waymo, at 0.014 nationally, is not 1.5 times worse than the professionals (as the report claims), but about eight times better.
To be sure, accurately assessing the risk of for-hire vehicles against the citywide average (including speeding teenagers, drunk drivers, and motorcycles) is a challenging task. Professional drivers have higher reporting requirements for accidents, which might put them at a disadvantage against all drivers in this comparison. It’s plausible that professional drivers may indeed be slightly safer when measured against some categories of injury. But Waymo easily clears this benchmark as long as comparable statistics are used. While assessing the true impacts on mortality will take more time, we can still reason from the substantial improvements in milder accidents that safety is getting better.
The report’s second argument is that robotaxis will worsen congestion. While this may happen, the arguments used here are similarly tenuous.
The comparison is between Waymo’s fleet, which drives empty 42 percent of the time, versus 33 percent for New York City’s Uber and Lyfts. However, this compares Waymo’s operations in a sprawling, lower-density service area against one of the densest ride-hail markets on earth. “Deadheading,” as it’s called, is the result of trip density, not whether a person or a robot is driving the car. If we deploy Waymo’s fleet in Manhattan, we will see the share of empty miles fall.
Robotaxis might even improve the situation. The report finds that yellow taxis are empty 59 percent of the time, and that they drive empty for 1.43 miles for each mile that they have a passenger. By that logic, replacing yellow taxis with robotaxis (even at the utilization rate they achieve in places like California) would reduce miles travelled by nearly a third. This finding highlights an unintended benefit of ride-hailing technologies like Uber, Lyft, or Waymo: by reducing the need for yellow taxis to circle the streets to find passengers, they can substantially reduce congestion.
While its streets are indeed safer than those of other major cities, New York still recorded 205 traffic deaths last year and tens of thousands of injuries, overwhelmingly caused by driver error. Robotaxis would lower this rate, if not to zero, then to something much closer to it.
Waymo’s technology isn’t perfect and needs additional testing on city roads and improvements over time. However, state law still requires a human driver behind the wheel, and Waymo’s modest safety-driver permit expired in March, with no follow-up framework. And the local policy conversation is being shaped by reports, such as the one from Open Plans, that paint a misleading picture of Waymo’s true safety record. Albany simultaneously demands that Waymo provide a statistical certainty of safety that is only possible after many billions of miles while prohibiting Waymo from actually producing that record.
Some of the report’s recommendations are reasonable and should be supported. Requiring public-data reporting from operators is reasonable, and the city should provide TLC and NYPD crash data using consistent definitions to provide clearer benchmarks. But what New York has chosen instead is a ban through paralysis. New Yorkers should not let three crashes in the rest of the country determine the future of their streets.