Tesla has published its first Robotaxi Safety Report, a page on tesla.com that counts the miles its robotaxis have driven with no safety monitor in the car and sets them against every collision it reported to the US road-safety regulator, NHTSA. It went up on 21 October 2026, hours before Tesla's third-quarter results. It also lists four cities as "coming soon": Las Vegas, Phoenix, Jacksonville and San Antonio, without dates. Not a Tesla App reported the page first; TeslAnt read it through an Internet Archive copy made at 18:51 UTC.
3.1 million miles, two-thirds of them in Austin
From 22 January to 18 October 2026, the fleet drove 3,112,541 miles with nobody from Tesla aboard, in six cities:
| City | Miles without a safety monitor | Share |
|---|---|---|
| Austin, Texas | 2,118,015 | 68.0% |
| Dallas, Texas | 608,130 | 19.5% |
| Houston, Texas | 225,964 | 7.3% |
| Orlando, Florida | 61,053 | 2.0% |
| Tampa, Florida | 53,117 | 1.7% |
| Miami, Florida | 46,262 | 1.5% |
The pace is the news. Tesla announced one million unsupervised miles at the Cybercab launch on 3 September, and its Q3 letter, published the same evening, put the total at "nearly two million" at the end of September. The page's weekly data show about 558,000 miles added in the week of 12–18 October alone, against 334,000 in the week of 28 September–4 October. The Bay Area service, where an employee sits in the driver's seat, is not included.
The zeros, and what they count
Tesla reports no one injured by a Robotaxi, no Robotaxi airbag deployments and no pedestrians or cyclists hit. Its downloadable list of reported collisions shows what sits behind those figures: 32 crashes between 23 May and 18 October, 30 of them with property damage only. In 13, the robotaxi was stationary. Fifteen involved a fixed object such as a dumpster, a kerb or a hanging chain, and three happened while a remote operator was driving or recovering the car, cases TeslAnt examined last week. That is one reported collision every 97,000 miles or so, by TeslAnt's arithmetic. The list differs from NHTSA's public file of 36 Tesla reports: it leaves out crashes with a safety monitor aboard, and it includes October crashes that NHTSA has not yet published.
Two entries mention a hospital. On 26 September in Dallas, a robotaxi drove over a pothole in a work zone at about 29 mph, and one of its two riders "alleged they sought medical care"; Tesla says its investigation found no injury. On 10 October, also in Dallas, an SUV was pushed into a robotaxi waiting at a stop sign after colliding with another car, whose driver was taken to hospital. Neither counts, because Tesla's injury figure covers only collisions in which the Robotaxi made first contact.
The human benchmark
Beside each zero, Tesla shows what human drivers "could have caused" over the 2.95 million Texas miles: 9.1 injury collisions, 4.1 airbag deployments and 0.52 collisions with a pedestrian or cyclist. The figures come from 2024 police records in Austin, Dallas and Houston, with the injury number scaled up for crashes never reported to police. The comparison is not like for like, and Tesla's own methodology says so: police reports do not record which vehicle made first contact, so the human figures count every such collision.
The report does not cover traffic violations. NHTSA is still deciding whether to investigate complaints that robotaxis sped through school zones.
Four cities next
Las Vegas follows Nevada's August decision to let Tesla run up to 5,000 robotaxis in Clark County. Jacksonville would be Tesla's fourth Florida city and San Antonio its fourth in Texas. Our July report on the Miami launch listed San Antonio among the cities already served; Tesla's page shows it is not.
What it means in Europe
For European owners, nothing on the page applies yet: FSD in Europe is the supervised version, now approved in nine EU countries. What the page offers is the first robotaxi safety record Tesla has published itself, with a stated method and a crash list that can be checked against NHTSA's data.