Why Tesla’s AI trainers don’t trust its self-driving tech

Added: 2026-06-03

[00:00:00]Big promise Elon Musk has made about full [music] self-driving technology is that it will soon turn every Tesla on the road into a driverless vehicle. It's a big reason why the company is valued about [music] $1.5 trillion, five times that of any other automaker.

[00:00:13]We found after talking to 10 former Tesla employees who helped train its self-driving software >> [music] >> that Tesla's technology is still far off from what Musk is promising. Seven of those employees said they wouldn't trust [music] the software to drive them anywhere. While at Tesla, the employees reviewed videos showing Teslas equipped with a full self-driving [music] software striking dogs, cats, and wildlife like deer, as well as close calls with pedestrians including [music] children. Some saw Teslas using the software fail to pull over for approaching emergency vehicles, >> [music] >> fail to stop for school buses, and drive into a construction zone nearly striking workers. The [music] vehicles using the technology often sped. One former employee said vehicles using a full self-driving mode called Mad Max [music] would go 20 to 30 mph over the speed limit. Even though Tesla, according to the employees, knew about these deficiencies, Musk pushed forward with the launch of a robotaxi network in Austin, Texas [music] last June. Our reporting shows that employees worked to train the software on specific streets and intersections [music] so that the demonstration would go smoothly. They said this kind of training is impossible to do on [music] a wide scale. Yet Musk has said Tesla's technology works anywhere and doesn't require mapping or programming the software for it to operate in specific places. [music] Another interesting thing we found is that Tesla is using some fuzzy math to claim that its full self-driving technology is either 10 times safer or >> [music] >> more recently seven times safer than the average human driver. We found after talking to 11 traffic safety experts that [music] Tesla is comparing apples and oranges to do this math. It's counting only the worst of its crashes, the ones where airbags go off, and comparing them to a set of [music] federal data that includes more minor accidents. This distorts Tesla's safety record [music] and makes its technology look better than it actually is. By contrast, Waymo, the US leader in driverless taxis, [music] takes a more rigorous approach.

[00:01:54]It publishes its safety stats in peer-reviewed journals and says it adjusts for the specific roads and conditions [music] its vehicles actually operated. We sent Tesla a detailed list of questions for this report, but got no response.

[00:02:06]In the past, [music] Tesla's defended its full self-driving technology saying it could save lives and prevent injuries. The traffic safety researchers who spoke to Reuters said they viewed Tesla's claims about the safety of full self-driving [music] as marketing rather than a serious investigation into a critical safety issue.