A human taking control of a self-driving vehicle is supposed to provide an additional layer of safety. But new federal crash records reveal that Tesla’s remote operators have been involved in two more collisions while attempting to move autonomous vehicles through difficult situations.
The incidents occurred in Dallas and Austin, Texas, during August 2026 and were highlighted in an October 15 review of federal safety data. Both involved Tesla Model Y robotaxis moving at extremely low speeds, with no passengers aboard and no reported injuries.
The discoveries bring the number of documented Tesla robotaxi collisions involving direct remote control to five. Although the incidents were relatively minor, they raise questions about the reliability of human backup systems, transparency and the challenges of operating vehicles remotely.
The findings arrive as Tesla faces increasing scrutiny over its autonomous-driving technology.
Tesla’s Robotaxi Crash Records Now Include Five Remote-Operator Collisions
Tesla’s latest disclosures to the National Highway Traffic Safety Administration have brought fresh attention to the human operators supporting its autonomous vehicles. An October 15 investigation by automotive publication Electrek identified two additional collisions in which remote operators were controlling Tesla robotaxis when contact occurred. Those discoveries brought the total number of documented remote-control collisions to five. The broader federal database contained 36 unique Tesla automated-driving incident reports, compared with 22 in July. The increase reflects additional incidents and disclosures, not necessarily a corresponding increase in the company’s crash rate.
Three of the five remote-operator cases are explicitly identified in federal records through the classification “Remote (Commercial / Test).” The other two were established through previously disclosed descriptions of crashes that occurred in Austin. Importantly, five is a count of known collisions involving remote control, not the number of times Tesla employees have intervened in total. The distinction matters because an operator may successfully recover a stranded vehicle without generating a reportable crash. Federal records therefore establish that these collisions occurred, but they do not reveal how frequently Tesla’s remote intervention system operates successfully.
Dallas Collision Occurs at Just 0.4 Miles Per Hour
One of the newly identified incidents occurred on a residential street in Dallas in August 2026. According to Tesla’s federal report, identified as 13781-16678, a 2026 Model Y was stopped between a concrete mixer truck on its right and a parked sport utility vehicle on its left. The autonomous vehicle apparently needed assistance navigating the confined space. A remote support operator subsequently assumed control and attempted to move the Tesla forward. During that manoeuvre, the vehicle contacted the parked SUV at approximately 0.4 miles per hour, a speed considerably slower than normal walking.
No passengers were inside the Tesla, and the incident was classified as involving property damage only. There were no reported injuries. Although the impact was minor, the circumstances illustrate a practical challenge for autonomous taxis. An ordinary residential street can become difficult to navigate when large commercial vehicles and parked cars leave limited room to manoeuvre. Tesla’s report establishes that the remote operator was controlling the vehicle at the time of contact, but it does not publicly explain the operator’s complete visual perspective or why the manoeuvre was unsuccessful. Those unanswered details matter when evaluating the reliability of the recovery system.
Austin Robotaxi Strikes a Dumpster While Under Remote Control
The second newly identified collision happened in Austin, where another 2026 Tesla Model Y was travelling through an alley. Federal report 13781-16679 describes a narrow passage with a utility pole on one side and a dumpster on the other. A remote support operator was controlling the vehicle as it proceeded straight ahead. Despite moving at approximately one mile per hour, the Tesla made contact with the dumpster. Like the Dallas incident, the collision resulted in property damage only, with no passengers aboard and no injuries reported.
The Austin report is notable because it explicitly identifies the driver or operator type as remote commercial or test operation. That makes the human involvement clear rather than leaving investigators to infer who was controlling the vehicle. Together, the Dallas and Austin incidents show how even slow manoeuvres can become difficult when space is limited and nearby objects must be avoided precisely. The records do not establish whether camera coverage, communication delays or human judgment contributed to the contacts. They do, however, raise a reasonable question: how effectively can a person manoeuvre a vehicle through a confined environment when operating it from somewhere else?
The Earlier Houston Crash Happened During a Roadside Recovery
An earlier collision in Houston shows how a seemingly straightforward recovery can become complicated. In May 2026, a Tesla Model Y stopped on a dead-end residential street and required help getting out. According to federal report 13781-15395, remote assistance was guiding the car away from the dead end when it moved onto a grassy slope. As the operator continued the recovery, the vehicle struck a hidden tree stump. Tesla reported a pre-crash speed of two miles per hour, no injuries, no airbag deployment and no need for towing. The event was classified as property damage only.
The Houston report was particularly revealing because Tesla explicitly marked the driver-or-operator category as remote operation, the first such structured designation identified in its then-public robotaxi records. That is a different kind of information from simply recording that autonomous software had been engaged shortly before an impact. The car needed human help, a person offsite assisted with moving it, and contact occurred during that process. Still, the public record does not establish whether the stump was visible to the operator or identify the precise factors behind the manoeuvre.
Two Earlier Austin Crashes First Exposed the Role of Teleoperators
The earliest documented incident in this group occurred in Austin in July 2025, shortly after Tesla began its robotaxi operations. An automated Model Y had stopped at the side of a street and would not proceed. An onboard safety monitor requested assistance, and a remote teleoperator took control. According to the subsequently disclosed report, the operator increased the vehicle’s speed, steered left, mounted a curb and struck a metal fence at approximately eight miles per hour. The safety monitor reported a minor injury but was not hospitalized; no paying passengers were aboard.
In January 2026, another Austin safety monitor requested help with navigation while a Tesla was stopped. A remote operator took over and proceeded straight ahead, striking a temporary construction barricade at roughly nine miles per hour. That collision scraped the front-left fender and tire, with no injury reported. Both accounts became public in May 2026 after Tesla stopped withholding the narrative portions of its older crash filings. They established that direct remote driving was not merely a theoretical emergency function: employees had already used it on public roads, with collisions occurring during two recoveries.
Tesla Allows Remote Driving Only Under Specific Speed Restrictions
Tesla has acknowledged that its remote-assistance employees can sometimes drive a robotaxi directly, although the company describes this as an exceptional last resort. In a March 26, 2026, response to U.S. Senator Edward Markey, Tesla said direct control may begin only when a vehicle is moving at no more than two miles per hour. Once authorized by the onboard system, the remote operator’s maximum permitted speed is 10 miles per hour. Tesla says staff ordinarily provide guidance or issue limited commands, including short forward or reverse movements, waypoint changes and system reboots.
The company identified remote-assistance operations in Austin, Texas, and Palo Alto, California, and said all its remote operators were U.S.-based employees. According to its response, workers must have held a valid American driver’s licence for at least three years, undergo background and drug screening, and complete training. Tesla also says sessions are logged and selected interactions audited. Those safeguards help explain why the company uses remote control as a recovery option. The five known collisions nevertheless make the performance of that last-resort system a legitimate subject of scrutiny.
Federal Crash Reports Reveal Who Was in Control, but Not Everything
The National Highway Traffic Safety Administration collects certain automated-vehicle crashes through its Standing General Order reporting program. For automated driving systems, an incident may be reportable when the technology was engaged at any point within 30 seconds before a qualifying collision. That reporting window matters: a record can involve both autonomous driving and a subsequent human takeover. The mere presence of a Tesla in the federal database therefore does not mean the automated software itself caused a collision, nor does it indicate that a paying passenger was aboard.
Public visibility has also changed over time. In May 2026, Tesla made narrative descriptions available for 17 older incidents after previously seeking confidentiality protections for those accounts. Some narratives described other vehicles hitting stopped Teslas, while others documented the company’s own vehicle striking obstacles. The detailed wording was crucial for identifying the original two remote-driving collisions in Austin. Of the five cases now highlighted, only three have the structured remote-operator label; the remaining two are identified by the narrative descriptions. Counting report IDs also requires care because updated versions of an incident can appear in the downloadable data.
The Newest Crashes Were Minor, but One Earlier Case Involved an Injury
The August incidents in Dallas and Austin were low-speed contacts with property, not high-speed impacts involving passengers. Tesla reported no injuries in either case. The Houston recovery also resulted in property damage only, with no injury or tow recorded. It would therefore be misleading to portray these three events as serious passenger crashes. A dented parked vehicle or damaged dumpster is materially different from an incident involving a pedestrian or a collision at highway speed, and the federal narratives should be read with that distinction in mind.
However, it would also be inaccurate to say that nobody was hurt across all five remote-control cases. The July 2025 fence collision in Austin resulted in a minor injury to an onboard safety monitor, according to the previously disclosed records and contemporaneous reporting by WIRED. The person was not hospitalized. The distinction underscores why separate incidents need separate descriptions rather than one sweeping claim about severity. Even when consequences are limited, companies still need to investigate what went wrong, how operators assessed the surroundings and whether similar contacts can be prevented as fleets grow.
Five Recorded Collisions Do Not Mean Every Remote Takeover Crashes
One interpretation of the data needs particular caution. The five known remote-control collisions were identified in a database of reported crashes, not a complete log of every time a Tesla employee remotely assisted a vehicle. If an operator moved a stuck robotaxi successfully and no reportable collision occurred, that intervention would generally not appear in the crash file. Consequently, describing the record as evidence that all remote takeovers end in collisions would confuse the selected cases with the entire population of interventions.
Tesla has not publicly disclosed the total number of remote-driving sessions needed to calculate a takeover-specific collision rate. That information was also a central concern of Senator Markey’s March 2026 review of remote-assistance practices: his office said all seven companies examined withheld intervention-frequency data. NHTSA separately warns that its crash counts are not adjusted for miles driven, fleet size or other exposure measures. The reported 36 Tesla incidents are consequently neither a standalone safety score nor proof of how the service compares with competing robotaxis. Reliable comparisons require transparent denominators, consistent definitions and careful analysis of circumstances.
Waymo Uses a Different Kind of Human Backup
Tesla’s approach differs from the remote-assistance model described by rival Waymo. In a February 2026 letter to Senator Markey, Waymo said its support agents provide information when a vehicle requests help, while the onboard driving system decides whether to accept or reject the suggestion. Waymo said those workers do not ordinarily steer, accelerate or brake the vehicle remotely. It described a narrow emergency tool allowing specially trained U.S.-based personnel to prompt a stopped car forward at two miles per hour with fixed steering angles, but stated that the tool had never been used outside training.
These contrasting designs illustrate an important question in autonomous transportation: who should be responsible for the final movement when a car becomes stuck? Tesla permits limited direct control; Waymo generally leaves real-time driving decisions to its software. Neither company’s explanation, by itself, proves its service is safer. Remote driving can be affected by network delays and incomplete visibility, while advice-only systems may have their own difficulties when software cannot resolve an unusual situation. Independent safety assessment must consider how each system behaves during actual recoveries, not merely how the manufacturers describe their procedures.
Regulators Are Pressing for More Transparency as Robotaxi Services Expand
The remote-control disclosures arrive amid wider regulatory attention to autonomous vehicles. In March 2026, Senator Markey’s office published findings from inquiries to seven developers, highlighting differences in operator training, communication delays and disclosure practices. Tesla told the senator that its remote-driving capability was tightly restricted, but it did not supply a public frequency for interventions. For people sharing streets with driverless vehicles, that missing information makes it harder to distinguish a rare recovery error from a more persistent operational weakness.
A separate federal development underscores the scrutiny. On October 9, NHTSA opened defect petition DP26010 to review allegations that 2026 Tesla Model Y and Cybercab robotaxis exceeded speed limits in active school zones. The petition cited three video links, but opening it was not a finding of a defect or a formal decision to investigate; the file reported no crashes or injuries associated with those allegations. That issue is separate from the five remote-operator collisions. Together, however, the cases show the questions Tesla must answer as its service develops: how often humans intervene, how safely they take control, and how clearly results are reported to the public.
