The Two Notable Tesla Robotaxi Crashes Were Caused by the Humans, Not the Robot

Here is the part of the Tesla Robotaxi story that nobody put in a headline. The two crashes everyone is talking about, the ones tucked inside the data Tesla finally unredacted for federal regulators on May 15, were not caused by the self-driving software glitching out. They were caused by the humans. The remote humans. The teleoperators who exist specifically so the car does not crash.

Read that twice, because the cat had to. A robotaxi is supposed to be the car that drives itself. The whole pitch, the entire reason the word “autonomous” gets stamped on the marketing, is that the human is removed from the loop. And then, in the two incidents that stand out from the seventeen crash narratives Tesla submitted to the National Highway Traffic Safety Administration, the car was being driven by a person sitting in an office somewhere, and that person drove it into a fence.

What actually happened in Austin

The details are almost gentle, which makes them funnier. In July 2025, a Tesla Robotaxi in Austin got stuck. The automated driving system was stopped on a street and could not figure out how to move forward, the kind of small hesitation you would forgive a nervous teenager for. The safety monitor in the car asked Tesla’s remote assistance team for help. A teleoperator took over, gradually increased the speed, turned left, drove up over the curb, and made contact with a metal fence.

Then it happened again. In January 2026, another car, another stalled moment, another safety monitor asking for navigational help. A teleoperator took control while the system sat motionless and drove straight into a temporary construction barricade, scraping the front-left fender and tire. Both crashes were low speed. Nobody was hurt. There were safety monitors behind the wheel and no passengers onboard. Tesla lets remote operators pilot the cars as long as they keep it under 10 miles per hour, which means both of these fence encounters happened at roughly the pace of a determined jog.

So the score, in the cleanest reading of the situation, is robot zero, fences two, and the points were both scored by people. This is the autonomous future arriving exactly the way real life always arrives, sideways and slightly embarrassing.

The word “teleoperator” is doing a lot of heavy lifting

Here is where the cat wants to ask the question nobody seems to ask out loud. If a robotaxi needs a person in a remote office ready to grab the wheel, and a safety monitor physically sitting in the car, and that whole arrangement still produces a car climbing a curb into a fence, what exactly is the robot part?

This is not a gotcha. Remote assistance is a normal, sensible safety layer. Waymo uses it. The honest version of autonomy in 2026 is that the cars handle most of the routine driving and humans get pulled in for the weird edge cases, the construction zones, the double-parked delivery van, the moment the software simply does not know what to do. The teleoperator is the seatbelt. Nobody is mad at the seatbelt.

The awkward part is the gap between that honest version and the marketing version. The marketing version says the car drives itself. The data submitted to the NHTSA says the car frequently gets confused, raises a hand like a kid who lost the worksheet, and a stranger several miles away takes over and occasionally drives it into a barricade. Those are two very different products wearing the same name.

Why this is a familiar shape

The cat keeps noticing the same pattern in tech this year. The promise is full automation, no humans, the machine handles everything. The reality is a human standing just out of frame, sweating, ready to catch the thing when it stumbles. We wrote about Google quietly killing Project Mariner, the AI agent built to watch your browser and click things for you, because the gap between “it works” and “it ships” turned out to be wide enough to bury a product in. We watched an AI coding agent wipe a startup’s database in nine seconds and then confess in all caps like a cat caught next to a broken glass. The teleoperator crash is the same story with a steering wheel attached.

It is also a useful reminder that automation does not delete human error. It relocates it. The mistake that used to happen at the wheel now happens at a desk, mediated by a video feed and a network connection, made by a person who cannot feel the curb under the tires or hear the scrape of the fence. The human is still the weak point. We just moved the weak point somewhere with worse situational awareness and called it progress.

The pressure is real and so is the spin

Tesla did not volunteer this. The crash narratives were unredacted, meaning the information existed and was hidden until it could not be anymore. Seventeen Robotaxi crash reports went to the federal regulator, and the two teleoperator incidents are the ones that put fresh pressure on the company, because they are the ones that complicate the story Tesla likes to tell. A car that crashes because its sensors failed is a software problem you can patch. A car that crashes because a remote human drove it into a fence is a harder thing to spin, since the remote human was supposed to be the responsible adult in the arrangement.

None of this means robotaxis are doomed. Low-speed, no-injury, no-passenger fender contact is genuinely not a catastrophe, and the technology will keep improving. But it does mean the marketing language deserves the side-eye a cat reserves for a closed door. When a company tells you the car drives itself, the correct follow-up question is the one we have been learning to ask about every piece of software this year. The same skepticism we suggested for Android’s new feature that makes you wait ten seconds before you doomscroll applies here. Read what the thing actually does, not what the slide deck says it does.

The cat’s verdict

A robotaxi is a car with a confidence problem and a support hotline. Most of the time it drives fine. Some of the time it freezes, asks for help, and a person you will never meet takes the wheel from a building you will never see. Twice now, in the public record, that person has driven it into something solid at jogging speed.

The autonomous car was sold as the end of human error on the road. What arrived is human error with extra steps and a worse view. The cat is not opposed to robotaxis. The cat is simply opposed to pretending the humans left the room. They did not. They are in the other room, on a headset, about to meet a fence.