Tesla's New Robotaxi Safety Report Claims 3.1 Million Driverless Miles and Zero Injuries. Tesla Also Wrote the Rules It's Graded On.

Honest fence (read first): On Wednesday, Oct 21, Tesla put up a new page at tesla.com/robotaxi/safety titled "Robotaxi Safety Report." It says the Robotaxi fleet, driving with no safety monitor inside, covered 3.1 million miles from Jan 22 to Oct 18, 2026, with zero people injured, zero Robotaxi airbag deployments and zero pedestrians or cyclists hit. Those zeros are Tesla's own count, on Tesla's own definitions. Nobody outside Tesla has audited them, and this is not an NHTSA finding. Tesla's page also says it left two filed events out of the injury count, and its crash file still lists roughly 32 reported crashes (that count comes from EV's read of the file; the box couldn't download it from tesla.com).

Editorial Thesis

This is the most Robotaxi data Tesla has ever published, and that deserves credit. City-by-city miles, a written methodology, sources, and a download of every crash it reported to the federal government: none of that existed last week. But a safety report is only as strong as the yardstick. Tesla picked the yardstick, and the zeros only mean what Tesla's definitions say they mean. The useful move for a reader is to keep the miles (strong evidence) apart from the zeros (a claim built on choices) and apart from the human comparison (a rough benchmark, counted differently).

Honest Verdict

Proven (on the page, with sources): Tesla's Robotaxi fleet has driven about 3.1 million no-monitor miles, mostly in Texas, and the growth since late August is steep. Claimed: zero injuries, zero airbags and zero pedestrian or cyclist hits, counted only when the Robotaxi "made first contact," plus two events Tesla decided not to count. Not proven: that Robotaxi is safer than human drivers by the margin the page implies. The human numbers count every crash no matter who hit whom, and Tesla says so itself.

What the page actually says

Everything in this section comes from tesla.com/robotaxi/safety, read at about 5:05 AM ET on Thursday, Oct 22.

• Miles: "3.1 M Miles Driven Autonomously, As of October 18, 2026." A separate live counter read 3,364,582 when we checked. Tesla says that counter is an estimate at the fleet's current pace, not a reported figure.
• By city: Austin 2,118,015 - Dallas 608,130 - Houston 225,964 - Orlando 61,053 - Tampa 53,117 - Miami 46,262. Those add up to 3,112,541 (our math). The three Texas cities alone are 2,952,109 miles, about 95% of the total.
• Conditions: 862,000 miles after dark, 113,000 in rain and 575,000 in weekday rush hour. Tesla calls the rain figure "a minimum" because it only counts hours when the local airport weather station reported rain.
• Results: "People Injured by Robotaxi" 0. "Robotaxi Airbag Deployments" 0. "Pedestrians or Cyclists Hit" 0.
• Human comparison: "Human drivers could have caused 9.1 collisions with injuries," 4.1 with airbag deployments and 0.52 with pedestrians or cyclists "in the same number of miles."
• Coming soon: Las Vegas, Jacksonville, Phoenix and San Antonio. No dates.

The page doesn't list the San Francisco Bay Area, where Tesla's service still runs with a safety driver behind the wheel. That's consistent with the page's own scope: no-monitor miles only.

How the zeros are counted

This is the part worth reading slowly. Tesla says the report "includes every collision Tesla reports to NHTSA under its Third Amended Standing General Order 2021-01" while Robotaxi drives with no monitor in the car, "including when a remote operator is assisting or has taken control." So far, that's a strong scope. It uses the federal reporting rule, and it includes remote driving.

Then come the definitions:

• "People Injured" counts anyone hurt "in a collision in which Robotaxi made first contact." If another car hits the Robotaxi and someone gets hurt, that doesn't count.
• "Airbag Deployments" counts Robotaxi's own airbags only.
• "Pedestrians or Cyclists Hit" again counts only first-contact collisions.

And Tesla says it left out two events that were filed with NHTSA:

1. A multi-vehicle crash in which another vehicle hit an SUV and pushed it into a Robotaxi "as it waited at a stop sign." Tesla says nobody in the Robotaxi or the SUV was hurt, but an occupant of the other vehicle went to the hospital. An airbag also deployed in another vehicle.
2. A rider "claimed injury after a Robotaxi drove over a pothole, but investigation found no injury." That investigation is Tesla's.

Both explanations may be fair. A Robotaxi sitting at a stop sign isn't at fault for being shoved, and a claim isn't proof of injury. But both calls were made by the company being graded. That's the core claim-vs-proof problem with any self-published safety report.

The human benchmark, in plain English

Tesla built its "could have caused" numbers from Texas police crash reports (TxDOT CRIS, 2024) for Austin, Dallas and Houston, divided by passenger-vehicle miles on those cities' surface streets. It then adds about 32% to the injury figure for crashes never reported to police, citing a 2023 NHTSA study (Blincoe et al.).

Tesla is upfront about the catch: "Police reports do not record which vehicle made first contact, so the human injury and pedestrian or cyclist figures count every such collision, whichever vehicle made first contact."

So the Robotaxi side counts only crashes where Robotaxi hit first. The human side counts every crash. That isn't like-for-like, and it tilts the comparison in Tesla's favour. By how much, nobody can say from the page.

Two more differences matter. The human figure covers all city surface-street driving, while Robotaxi drives inside geofences Tesla chose, with hours it controls. And the benchmark uses 2024 crash data against 2026 driving.

EV, an independent EV site, worked out that Tesla's benchmark is about 3.1 injury collisions per million miles. Waymo uses about 3.25 per million for human drivers in Austin on its own safety hub (EV's comparison). So Tesla's human number is in the same range other operators use. The definitional gap on the Robotaxi side is the bigger issue.

What the crash file shows (secondary read)

Tesla offers a download of "all SGO reportable events." Our box couldn't fetch it (tesla.com returns 403 to our server), so we're leaning on EV's Oct 21 analysis here, and we label it that way:

• 32 crashes from May 23 to Oct 18, all but two reported as property damage only.
• In all but one, the Robotaxi was moving at 6 mph or less, or stopped.
• In 16, another vehicle struck the Robotaxi, often from behind at a light.
• Six involved low-hanging chains strung across parking-lot entrances. Three happened under remote control.
• 17 of the 32 already appear in NHTSA's public data, which the agency releases a month or more after filing. The other 15 are new to the public.

That's a much more useful picture than the zeros. Most of the reported crashes look like low-speed scrapes and getting rear-ended. Chains across lot entrances are a real, specific weak spot. Remote-control crashes are worth watching as the fleet grows. And "zero injuries" never meant "zero crashes."

The miles are the strongest part

The mileage chart is where the page is most convincing, because it's hard to fake and easy to check over time. EV's read of the weekly data puts the fleet at about 937,000 cumulative miles by the week of Aug 24, then about 558,000 miles in the single week starting Oct 12, with 437,800 of those in Austin. Roughly 70% of all no-monitor miles came in the seven weeks after late August. That lines up with the Cybercab launch in Austin and the Texas DMV roster: our Robotaxi Tracker read 349 Cybercabs and 454 Model Y on Tesla's Texas AV roster as of Oct 21.

The volume also matters for statistics. At Tesla's own benchmark, an average human would be expected to have about nine injury crashes over these miles. Seeing none in the first-contact category is meaningful if the categories matched. That's why the definitions matter so much: the data is getting big enough to say something, as long as it's counted the same way on both sides.

Claim vs proof

Why it matters now

Tesla has an NHTSA Special Order on how it self-certified the Cybercab due Oct 30, and NHTSA is also reviewing school-zone complaints. A public safety page helps Tesla tell its side, and the SGO download gives regulators and reporters a head start. It also sets a bar. Now that Tesla publishes this, a quiet month or a changed definition will be noticed.

What to watch next

• Whether Tesla updates the page on a regular schedule, and keeps the same definitions when it does.
• NHTSA's next SGO data release: do Tesla's file and the federal data still match?
• Any outside analysis that applies one counting rule to both Robotaxi and human crashes.
• Dates or permits for the four "coming soon" cities.
• Tesla's Oct 30 answer to NHTSA on Cybercab certification.

Video note

CBS Austin's Oct 15 report (pW3rbnGsTeI, 1:54) shows Cybercabs and Model Y Robotaxis on Austin streets and the "safety questions following close behind." It predates the safety page by six days, so it's context for the fleet the page describes, not coverage of the report itself.