The TL;DR
A Reuters investigation found that 7 out of 9 Tesla data labels will not mount on the FSD. They were used to seeing the show speed up and fail on camera.
Reuters interviewed nine Tesla data labels and a former self-driving engineer about their thoughts on Tesla’s Full Self-Driving mode. Seven out of nine data experts said they would not drive a Tesla running on FSD. One said they would never ride a Tesla robot”if you pay me the king.“
“We have all seen it fail,” one insider told Reuters. The former self-driving engineer admitted: “Definitely don’t trust Elon on this one.“They were referring to Musk’s announcement that Tesla cars are ready “it is safe and unguarded” ride.
The data labels task was to compile hours of FSD video and train the vehicle’s software to avoid past mistakes. They had direct access to the driver’s identity database. At least five told Reuters they regularly saw clips of Teslas driving above the speed limit while working on the FSD.
The issue of acceleration was taken very lightly by engineers and managers. Edge-case problems, such as unusual road conditions or unusual lighting conditions, have received more attention. Standard speed, which affects all drivers and all roads, was not significant.
💜 for EU tech
The latest talk from the EU tech scene, a story from our genius founder Boris, and some incredible AI art. It’s free, every week, in your inbox. Register now!
The investigation comes as Tesla has expanded FSD availability to new markets. Tesla confirmed the availability of FSD in China last week, although it is not yet clear whether ordinary consumers can still use the system. The FSD (Supervised) system is classified as Level 2, which requires constant driver attention. A completely autonomous unsupervised version is being tested only at a network of robots in Austin, Texas.
Recent months have produced a series of FSD-related incidents. Teslas working on FSD have driven over lakes, bridges, and in the path of oncoming trains. These are the incidents that reached the media. The evidence of the data labels suggests that the internal recording contains a very large catalog of failures.
The gap between Musk’s claims and the system’s performance has been an ongoing problem. Musk has promised fully autonomous driving repeatedly since 2016. Each deadline passed without delivery. The company’s robotaxi service in Austin operates in a geofenced area with remotely located security drivers.
Waymo’s flood-related shutdown this month showed that even the most advanced autonomous driving systems have ways to fail in normal situations. Tesla’s approach is very different from Waymo’s: camera-only vision versus multi-sensor integration, and consumer car intended for autonomy versus purpose-built robot.
Data label evidence is important because these operators are very close to raw performance data. They don’t see marketing materials or earnings call projections. They see hours of video showing exactly how the software behaves on public roads. Seven of the nine couldn’t get on board the product they helped build.
Tesla did not respond to Reuters’ request for comment. The company has previously stated that FSD (Supervised) requires driver supervision and its safety statistics show that the system outperforms human drivers per mile. A former engineer Reuters interviewed disputed those figures.
The investigation raises a question that Tesla’s maintenance and sales do not address: if the people who train the AI don’t trust it, why should the people who ride in it?
