Tesla's Robotaxi Crash Filings Now Show Five Remote Takeovers and Five Collisions

Two more Tesla Robotaxi crashes in which a remote operator, not the autonomous system, was driving have appeared in federal data, Electrek reported on Oct. 15. That makes five remote takeovers visible in Tesla's crash reports, and every one ended in contact with an object.

Both new crashes happened in August and were published in the National Highway Traffic Safety Administration's September release of Standing General Order data, the filings that automated-driving companies must submit after a crash. Tesla now has 36 unique Robotaxi incident reports on file, up from 22 when Electrek last counted in July.

What the two August filings say

In Dallas, report 13781-16678 describes a Model Y stopped on a residential street between a concrete mixer truck and a parked SUV. Tesla wrote that "the remote support operator remotely took control and while moving forward, the vehicle made contact with the SUV at approximately 0.40MPH," according to Electrek's reading of the filing. No passengers were aboard.

In Austin, report 13781-16679 has a remote operator "in control" of a Model Y going straight through an alley with a utility pole on one side and a dumpster on the other. It hit the dumpster at about 1 mph. Both crashes were property damage only, and both carry Tesla's structured code for operator type, "Remote (Commercial / Test)."

That code is the same one Tesla first used for a May crash in Houston, when a Model Y was being driven out of a dead-end road and hit a hidden tree stump at 2 mph. Three of the five are now formally coded as remote operation.

Date	City	Speed at impact	Object hit
July 2025	Austin	8 mph	Metal fence, after a curb
January 2026	Austin	9 mph	Construction barricade
May 2026	Houston	2 mph	Tree stump
August 2026	Dallas	0.40 mph	Parked SUV
August 2026	Austin	1 mph	Dumpster

The first two rows surfaced only because Tesla unredacted its crash narratives in May, after earlier arguing that disclosure would cause it financial harm. They carry no remote-operator code in the structured data, so they are known from the narrative text alone.

1. Austin, January 20269 mph
2. Austin, July 20258 mph
3. Houston, May 20262 mph
4. Austin, August 20261 mph
5. Dallas, August 20260.4 mph

Source: Electrek reporting on Tesla NHTSA Standing General Order filings, Oct. 15 and July 20, 2026

Five for five has no denominator

The number that matters here is the count of remote takeovers that did not end in a crash, and nobody outside Tesla has it. Electrek wrote that Tesla does not disclose how often remote operators intervene, so no rate can be calculated. Five collisions in 36 reports says nothing about the frequency of takeovers, which could be 5 or 5,000.

The filings are also a biased sample. A crash report exists because a collision happened; a takeover that ended with the car rolling clear leaves no filing. "Five for five" is therefore partly an artefact of what the reporting rule captures. What it does show is a pattern in the narratives: the autonomous system gets stuck, a human drives the car out, and the car touches something while the human is at the controls.

Severity is low. Speeds range from 0.40 mph to 9 mph, nobody was hurt in the crashes Electrek describes, and Tesla's robotaxis run on surface streets at limited speeds.

Where Waymo draws the line

Waymo has told a US senator it handles the same situation differently. In a reply to Sen. Ed Markey, reported by Reuters syndication on Feb. 17, Waymo said its remote assistance personnel 'do not directly control, steer, or drive the vehicle.' It runs four remote assistance centers, in Arizona, Michigan and two Philippine cities, with about 70 agents on duty at any time.

There is one exception. In rare cases, US-based staff on its Event Response Team can prompt a stopped robotaxi to move forward at 2 mph (3 kph) for a short distance to leave a travel lane. Waymo said this has not happened outside training.

Both companies built a way to move a stuck car. Waymo capped its version at a crawl and says it has never used it on a public road; Tesla's filings show remote drivers at up to 9 mph, in two of the five cases.

The remote-operator filings sit beside other open questions about Tesla's fleet; see our report on the school-zone petition filed with NHTSA, and our piece on Doug Field's role at Mind Robotics, the Rivian-backed robotics startup that hired the former Ford EV chief.

What would change this read

A disclosure of takeover counts per 1,000 miles would settle it. If Tesla's remote operators intervene thousands of times with five collisions, the record is better than it looks; if they intervene a few dozen times, it is worse. Tesla has not published that figure, and the next NHTSA data release will show whether the count of remote-coded crashes moves past five.