The Data on Driverless · Through June 2026A data story · September 2026
The car with no driver is the safer one.
Waymo’s self-driving cars had 3 bad crashes. People driving the same distance, in the same places, would have had about 58.
58expected with human drivers
3with Waymo
Each car is one vehicle in a crash where someone was seriously hurt or killed. The human figure is an estimate. What this does and doesn’t show
271M
miles with nobody driving
Roughly the Moon and back, 568 times over.
95%
fewer serious crashes, mile for mile
The plausible range runs from 83% to 99%.
6 in 10
of Waymo’s crashes happened while it was standing still
Stopped at a light, in traffic, or parked.
Sources: Waymo’s published safety data, and the crash reports it files with the federal government (NHTSA).Limitations and method below ↓
01Where it drivesWaymo mileage by map cell, decoded
Five metro areas, drawn to the same scale.
Each square is one S2 cell, roughly half a square mile, shaded by the rider-only miles driven in it. Together they account for all 271.3 million miles.
02The odometerWaymo data hub releases, Jun 2024 to Jun 2026
Almost four years to drive its first 22 million miles. It now drives that far in about six weeks.
Waymo’s rider-only record begins in September 2020. By June 2024 the fleet had logged 22.2 million miles. In the three months to June 2026 it logged 50.7 million, or about 557,000 a day. Each release covers slightly different areas, so the quarterly differences are approximate.
Rider-only miles by metro
Show the numbers
03Same miles, two driversWaymo against a location-adjusted human benchmark
The gap holds across most kinds of crash.
For each kind of crash, Waymo estimates how many human drivers would be expected to have over the same number of miles, weighted towards the neighborhoods where its cars actually drive. Each ring is one of those expected crashes. Filled dots are the ones Waymo actually had. The empty rings are the gap between the two: not specific crashes that were prevented, but the difference in rates made visible.
58
Expected with human drivers
3
Recorded with Waymo
55
The gap
95% fewer · 95% CI 83–99%
Outcome
With Waymo
Expected, humans
Gap
Reduction
95% CI
Counts are vehicles, not people or separate collisions: a crash involving two cars counts twice, as it does in the human benchmark. The categories overlap, so they shouldn’t be added together. Serious injury or worse is the fairest comparison, because both sides take their injury severity from police reports. The fatal row is the weakest: it rests on 2 Waymo crashes, and under a simpler, county-wide benchmark its range runs from 98% fewer to 10% more.
04Standing stillFrom the crash reports Waymo files with NHTSA
When a Waymo crashes, it usually isn’t moving.
Waymo’s summary files don’t say what its car was doing at the moment of a crash. The reports it files with the federal government do. I matched 1,964 of Waymo’s 1,977 crash records to their reports and read off the car’s movement and speed just before impact. The other 13 were either missing or shared a report with a second Waymo, which made them ambiguous.
What the Waymo was doing
Waymo speed just before impact, mph
62%
of crashes happened while the Waymo was stopped or parked
23%
of crashes involved a hit to the Waymo’s rear
41%
of car-on-car and single-vehicle crashes changed the Waymo’s speed by less than 1 mph on impact. That is a measure of how hard it was hit, not how fast it was going.
How does a car crash when it isn’t moving?
Something else runs into it. Of the 1,210 crashes in which the Waymo was stopped or parked, the reports classify the collision like this:
31%
Front-to-rear
The usual shape of being run into from behind while waiting in traffic.
22%
Side-on
A glancing contact along the side, typical of a car passing too close.
20%
Reversing
A vehicle backing up, often out of a parking space or driveway.
13%
Head-on
Front-to-front contact, often a vehicle pulling in or turning.
These are the collision types recorded in Waymo’s data, not findings of fault. The descriptions are typical cases, not an account of every crash. The remaining 14% are intersection and chain-reaction crashes, crashes involving cyclists, motorcyclists or pedestrians, and other types.
05The worst onesSummarized from Waymo’s NHTSA reports
The most serious crashes, as Waymo reported them.
Three crashes in the data involved a serious or fatal injury. A fourth, involving two Waymos, is still waiting to be classified. In each, Waymo’s report describes another vehicle striking the Waymo. These are Waymo’s own accounts, and police findings may differ.
Nov 2023 · San FranciscoSerious injury
A driver ran a red light
The Waymo was crossing an intersection on green, alongside another car. A car running the red hit both, then struck pedestrians waiting on the corner.
Waymo: driving straight · 18 mph
NHTSA report 30270-6579
Jan 2025 · San FranciscoFatal
A speeding SUV hit a queue of stopped cars
The Waymo, with no one inside, was stopped in traffic. An SUV traveling at very high speed hit the car behind it and set off a chain collision involving at least six vehicles. One person and a dog were killed.
Waymo: stopped · 0 mph
NHTSA report 30270-9724
Sep 2025 · TempeFatal
A motorcycle hit it from behind
The Waymo was slowing to turn right and giving way to a pedestrian. A motorcycle struck its rear corner. The rider fell into the next lane and was hit by a passing car, whose driver did not stop.
Waymo: turning right · 8 mph
NHTSA report 30270-11713
Jan 2026 · TempeUnder review
A red-light runner hit two stopped Waymos
Both Waymos were waiting at a red light. A speeding car ran it, hit another car, then hit both Waymos. A later police document described one driver’s injuries as life-threatening. Waymo says that if this is counted, its serious-injury rate rises from 0.011 to 0.018 per million miles (both Waymos count), against 0.213 for human drivers.
Waymo: stopped · 0 mph
NHTSA report 30270-13817
06LimitationsWhat this data can and can’t tell you
Where these numbers stop.
One company, five metro areas, six years. The result is strong, but it is narrower than it might first appear.
Waymo built the comparison. The methods are peer-reviewed and the data is public, but this is not independent research. I reproduced Waymo’s figures; I did not rebuild its human benchmark.
The human side is an estimate. It comes from police records, adjusted for where Waymo drives. It is not matched to the same routes, drivers, times of day or weather.
The worst outcomes are rare. Three serious crashes and two fatal ones leave wide ranges, and the fatal result is not statistically clear under every benchmark.
It covers a narrow slice of driving. Mostly city streets in five mild-weather metro areas. It says nothing yet about snow, rural roads, or other companies’ cars.
Being in a crash is not the same as causing one. Both sides count every crash, whoever was at fault, and being stopped at impact is a strong hint, not a verdict. Critics also argue that driverless cars brake more cautiously than people expect, which may make them likelier to be hit from behind.
Waymo has had failures of its own. It recalled software after its cars passed stopped school buses in Austin (Dec 2025), and a Waymo struck a child at low speed near a Santa Monica school (Jan 2026). Federal regulators opened investigations into both.
07Method and sources
How this was built.
Waymo Safety Impact Data HubThe September 24, 2026 release, with the corrected files Waymo republished on October 2, covering data through June 2026: miles by county, the 1,977 crash records behind the figures, the benchmark comparisons, and mileage by map cell. All comparisons use Waymo’s location-adjusted benchmark, its primary method. The headline figures are reproduced directly from Waymo’s published comparison table.
NHTSA crash reportsThe Standing General Order files for automated driving systems: the 2021–2025 archive and the current file (released August 17, 2026), using the latest version of each report. 1,974 of Waymo’s 1,977 records match a report. Ten records that share a report with a second Waymo were set aside, leaving 1,964 for the movement and speed analysis. This set includes a handful of freeway crashes and locations outside the five metro areas.
GrowthMileage from each quarterly data hub release since June 2024, recovered from the links in Waymo’s release notes.
MapsMap-cell IDs decoded to coordinates with my own implementation of Google’s S2 geometry, at level 13. Cell sizes vary slightly with latitude.
Outside researchThe Insurance Institute for Highway Safety (July 2026) found 68% fewer police-reportable crashes for the driverless Waymos it studied. Different measure, different design, same direction.
ChecksThe key figures are reproducible from the raw files with the analysis scripts, and were cross-checked in an independent review.
Why there are no robotaxis in New York.
Waymo tested its cars in New York for seven months, with no crashes reported. Then its permits ran out, and they weren’t renewed.
0
collisions during NYC testing
Up to eight cars with safety drivers, according to the city. A small trial, and not a measure of how a driverless service would perform.
~100K
NYC drivers whose jobs unions say are at stake
Taxi, rideshare and transit unions led the opposition.
5%
of Americans have ever ridden in one
And 71% say they would be uncomfortable doing so (Pew, 2026).
A summary of public records and reporting, kept separate from the safety analysis. Sources and dates below.Back to the safety data
01What happened in New YorkAug 2025 to Sep 2026
Seven months of testing, one withdrawn proposal, and a lapsed permit.
New York law requires a licensed driver with a hand on the wheel. A driverless service therefore needs the law to change, and so far it hasn’t.
Aug 22, 2025
New York City issues its first self-driving test permit, to Waymo. Up to eight cars, with safety drivers, in Manhattan below 112th Street and Downtown Brooklyn. Mayor Adams calls it an “innovative way to safely move our city forward.”
Aug 25, 2025
Driver groups rally against it outside the governor’s Midtown office. The Independent Drivers Guild says New Yorkers are being made “lab rats for Silicon Valley investors.”
Jan 13, 2026
Governor Hochul proposes allowing driverless ride-hail across the state, New York City excepted. The same day, Mayor Mamdani says his policy will focus “on the drivers who are here alongside me.”
Late Jan 2026
A Siena poll finds robotaxis unpopular. About 60% say they would feel unsafe, and about a quarter of city residents support them. The poll was paid for by the Black Car Fund, an insurer for for-hire drivers.
Feb 19, 2026
Hochul withdraws the proposal. Her office: “it was clear that the support was not there to advance this proposal.” Waymo says it is “disappointed” and will work with the Legislature.
Mar 31, 2026
Waymo’s city and state test permits expire without renewal, and testing stops.
Jul 20, 2026
A Hunter College and Open Plans study finds Waymo less safe than New York’s for-hire drivers. Its safety analysis is later retracted over errors in the data and definitions.
Sep 20, 2026
A Waymo-funded group, A Better Road Forward, begins running TV ads urging Albany to adopt a statewide framework. Its board includes the former Council Speaker Christine Quinn.
In Albany, bills on both sides stalled in committee in 2026: A4901, which would allow driverless cars, and S9038, the “Take the Wheel Act,” which would require a human driver in every for-hire car (status checked September 26, 2026). I found no report of Waymo reapplying for a permit, though that is not proof it hasn’t.
02The argumentsPut as each side puts them
Both sides have a serious point.
The concerns
Jobs. About 100,000 New York City drivers, many of them immigrants, and many still recovering from the taxi-medallion debt crisis.
Spillover. The Taxi Workers Alliance opposed even the upstate-only plan: “there is no practical boundary in this industry.”
A local comparison. The five-metro benchmark doesn’t show how Waymo would compare with New York’s taxis, ride-hail drivers or public transport.
Traffic and curb space. Empty cars circling between fares could add miles to already crowded streets.
Accessibility. Some disability advocates ask how riders who need help boarding, or a wheelchair-accessible car, would be served (Gothamist).
Accountability when no one is at the wheel.
The case for
Safety. Far fewer serious crashes than the human benchmark (see the safety data).
Independence. Rodney Stanford of the American Council of the Blind of New York, which partners with Waymo, says robotaxis “will obey the laws.”
No drunk or distracted driving. Mothers Against Drunk Driving supports them, and discloses Waymo funding.
Feeling safe. Waymo says women riders in particular feel safer than in conventional ride-hail.
Keeping pace. Waymo notes that more than half of US states have legalized fully driverless cars.
03Beyond New YorkStates, cities and Washington
Robotaxis run where state law makes room for them.
Waymo offers paid rides in 15 US metro areas, mostly in states whose laws explicitly allow driverless cars and prevent cities from banning them. Permission to test, to run without a driver, and to charge fares are separate things, and the absence of a service is not by itself evidence of a ban.
Where
Status as reported, Sep 2026
California
Operating. Regulators approved Waymo in 18 counties in August 2026, freeways included. Since July, police can ticket the company.
Texas
Operating. A state authorization has been required since May 2026. Tesla began paid Cybercab rides in Austin on September 3.
AZ, GA, FL, TN, NV, CO
Operating in Phoenix, Atlanta, Miami, Orlando, Tampa, Nashville, Las Vegas and Denver.
Washington, DC
No driverless service yet. A May 2026 bill would allow it from 2028, with a per-mile fee.
Illinois, New Jersey
No driverless service yet. Bills are pending; New Jersey’s would require lidar and radar.
Boston, Massachusetts
No driverless service yet. Proposals would require a human operator.
Federal
Regulators are easing the way, including the first exemption for a car without a steering wheel (Zoox, July 2026). Congress has not passed a comprehensive self-driving law.
Compiled from news reporting and official announcements. Rules change quickly; check the relevant state agency for current status.
04What people thinkNational and city polls
Most Americans are wary. Those who have ridden in one are far more at ease.
Haven’t ridden in one
5%
Have ridden in one
40%
Share who feel extremely or very comfortable riding in a driverless car. Pew, February 2026, 5,119 adults; riders are a small subset. People who choose to ride may have been more open to begin with.
Wary, nationally71% of Americans say they would be uncomfortable riding in a driverless car, and only 5% have tried one (Pew, 2026).
Experience and trust move togetherA USC study that followed the same people over time found that trust rose more among those who went on to ride in one. It is suggestive rather than conclusive, since people choose whether to ride.
San Francisco has warmedSupport rose from 44% in 2023 to 67% in 2025, after rides opened to the public, in polling paid for by GrowSF, a tech-backed group. An SF Chronicle poll in 2026 was more mixed: 42% positive, 18% negative, 39% undecided.
New York is coolAbout one in four city residents supported robotaxis, in a poll paid for by a driver insurer.