Showing Work: Why Tesla Published Every Robotaxi Crash It Reported to NHTSA

Tesla did something Wednesday it has never done before. It published a list of every crash its robotaxi fleet reported to federal regulators, explained in footnotes what the numbers mean and what they do not, cited the sources behind its human driver comparison and offered a download link so anyone can check the data themselves. The move is transparency at a level the autonomous vehicle industry has not seen from Tesla, and the timing matters.

The company's sworn response to NHTSA's audit of the Cybercab certification is due 30 October. That audit, opened the day public rides began in Austin, asks how Tesla determined a car with no steering wheel, no brake pedal and no gas pedal complies with federal motor vehicle safety standards. Publishing detailed safety data one week before that deadline is not a coincidence. It is strategy.

The safety report does two things at once. It gives the public the most complete account yet of how the fleet has performed, and it gives NHTSA a public benchmark against which to judge the Cybercab. Tesla is saying the data shows the car is safe, the methods are transparent and the results speak for themselves. The report is not a legal filing, but it sits in the same space. It is evidence, and Tesla made it public.

The zero-injury claim is the headline, but the footnotes matter more. Tesla defines what it counts as an injury, explains why two incidents were excluded, and breaks out what happens when the robotaxi makes first contact versus what happens when another vehicle hits it. The report includes crashes where a remote operator was driving, not just the autonomous system, and it excludes miles and collisions with a safety monitor in the car. Those are choices, and Tesla explains them.

The human driver comparison is careful. Tesla uses Texas police reports from the three cities where the fleet has driven 95 percent of its miles, adjusts for the roughly 32 percent of injury crashes that never get reported to police, and applies the collision rate per passenger-vehicle mile on surface streets in those cities. The comparison counts every collision a human driver was involved in, whether or not that driver made first contact, because police reports do not record which vehicle hit first. Tesla is not claiming its fleet is safer than the safest human drivers. It is claiming it is safer than the average.

The methodology section cites its sources. Federal crash data from Blincoe et al. 2023. Texas police reports from TxDOT CRIS. NHTSA's Standing General Order. Each claim has a reference, and each reference is public. Tesla is inviting scrutiny, not avoiding it. That is new.

The timing is also strategic in another way. NHTSA proposed a rule change that would remove some of the manual control requirements Tesla's audit is examining, but that rule is not final. Publishing safety data now, while the rule is pending and the audit is open, puts numbers in front of regulators who are deciding what standards should apply to vehicles like the Cybercab. The data says the car works without the controls. The report says Tesla has the evidence to prove it.

The download link is the tell. Tesla did not summarize the crashes or describe them in general terms. It posted a spreadsheet with every report the fleet filed with NHTSA in the period covered. Anyone can open it, sort by city or date or crash type, and see what happened. That level of access is rare in an industry that has faced criticism for keeping crash data close. Tesla is not keeping it close. Tesla is publishing it.

The report also serves another audience. Riders in the six cities where the service operates, and the four cities listed as coming soon, now have a public safety record to review before they book a ride. The live counter shows miles driven in real time. The city-by-city breakdown shows where the fleet is active. The conditions breakdown shows how many miles came after dark, in rain, in rush hour. The data is there, and it is public, and anyone deciding whether to trust a driverless car can read it.

The safety report is not a guarantee. Thirty-six crashes have been reported to NHTSA since June 2025, and the report includes them. Most were low-speed contact with stationary objects. None caused an injury under Tesla's first-contact definition, but they are crashes nonetheless. The network processes a million pixels per millisecond, but it still hits dumpsters, chains and potholes. The report does not hide that. It includes it.

What the report does hide, or at least not explain, is why Tesla is publishing this now and not six months ago or six months from now. The NHTSA audit deadline is the obvious answer, but there may be others. The fleet is about to expand to four new cities. Production is ramping. The Cybercab is heading to Dallas next month. Publishing safety data ahead of that expansion makes sense. It gives riders and regulators a baseline before the fleet grows.

The report is also a statement about what Tesla thinks the standard should be. The company is not waiting for NHTSA to publish crash rates or approve benchmarks. Tesla is publishing its own, with its own methods, and daring anyone to find a flaw. That is confidence, or it is pressure, or both. Either way, it is a move the industry has not seen Tesla make before. The company that has fought regulators, downplayed crashes and resisted disclosure is now publishing everything and citing its sources. The change is worth noting.

The safety site is live at Tesla.com/robotaxi/safety. The data covers 3.1 million miles, zero injuries under the first-contact definition, and a downloadable list of every crash the fleet reported. The methodology is explained. The sources are cited. The timing, one week before the NHTSA deadline, is no accident. Tesla is showing its work. The question is whether the work will be enough.