New federal filings shed light on robotaxi incidents
Tesla has released new details in a federal database about 17 robotaxi incidents that took place between July 2025 and March 2026, offering a clearer look at how its still-developing autonomous fleet has run into trouble. The disclosures are notable not only because Tesla had previously kept many specifics out of public view, but also because at least two of the reported crashes involved direct human action from remote operators.
According to the supplied source material, both of those incidents took place in Austin. In each case, a safety monitor was seated in the passenger seat and no paying passengers were in the vehicle. Both crashes happened at speeds below 10 miles per hour, but they still underscore a central tension in autonomous operations: the systems may be marketed as self-driving, yet humans remain essential to keeping them moving when edge cases appear.
What happened in the two remote-driving crashes
In one crash from July 2025, a safety monitor requested help from Tesla’s remote driving team after the vehicle stopped on the side of a street and would not proceed. A remote worker then drove the car up a curb and into a metal fence at 8 mph. Tesla reported that the monitor suffered minor injuries and was not hospitalized.
In a second case, from January 2026, a safety monitor requested navigation help from the remote team. The remote driver took control and drove the vehicle into a temporary construction barricade at 9 mph. Tesla reported damage to the robotaxi’s front left fender and tire but no injuries.
Both events are low-speed crashes, but they matter because they complicate the usual public framing around autonomous safety. The problem in these cases was not simply what the vehicle saw or failed to see on its own. It was also how human intervention was structured and executed once the system encountered a situation it could not cleanly resolve.
Why remote operations matter
The new details draw attention to a part of autonomous-vehicle operations that is rarely foregrounded in consumer marketing: remote assistance teams. The supplied reporting says all U.S. self-driving operators maintain these teams. But Tesla appears unusual because it more frequently allows remote workers to directly drive the cars.
That distinction is important. Other companies more commonly allow remote personnel to provide input or guidance to the vehicle software, which then decides whether and how to use it. The source notes that Waymo says specially trained workers can remotely drive its cars at up to 2 mph, but that the company said in February it had not used that function outside training. Tesla’s model, by contrast, appears to permit a more active takeover role.
This creates a different kind of safety question. Remote driving is not simply “human backup” in the abstract. It depends on communications links, situational awareness, camera feeds, interface design, and the operator’s understanding of the vehicle’s environment. If any of those are incomplete or delayed, a remote intervention can introduce a new failure mode precisely when it is supposed to reduce risk.
A broader test for autonomous transparency
Tesla’s disclosures also matter because they push more detail into the public record. For more than a year, the company had shielded specifics about robotaxi crashes from wider view. Now that those details are emerging, the focus is shifting from whether incidents happened to what kind of incidents they were and who was involved when they occurred.
The distinction is especially relevant for regulators, researchers, and the public. A low-speed collision caused by a remote human operator is operationally different from a failure of pure onboard autonomy, but it is still part of the robotaxi safety picture. The customer promise is not just that the car can drive itself under ideal conditions. It is that the whole system, including human oversight, can safely handle the moments when things go wrong.
The human layer is part of the product
Tesla’s newly disclosed incidents make one point especially clear: autonomous vehicles are not judged only by their software stacks. They are judged by the full chain of oversight around them, including safety monitors, remote support personnel, escalation procedures, and the rules that govern when humans can take over.
The Austin crashes may not have involved severe injuries, but they offer a useful window into the operational reality of robotaxis. Behind the driverless promise sits a network of humans making decisions at critical moments. When those decisions go wrong, the result is still a crash, and the accountability question becomes harder rather than easier.
That is why Tesla’s new filings matter. They do more than document minor collisions. They show that the human backstops inside autonomous systems are not peripheral details. They are part of the system itself.
This article is based on reporting by Wired. Read the original article.
Originally published on wired.com
