Tesla's FSD safety claims challenged as data labelers refuse to ride

A Reuters investigation found that Tesla’s safety statistics for its Full Self‑Driving (FSD) system are based on flawed comparisons. The company has promoted figures such as “10‑times safer” and “seven times safer” by counting crashes with airbag deployment against a broader set of crashes that include tow‑truck removals, and by comparing its relatively new fleet to an older average U.S. vehicle fleet. Researchers say a more appropriate metric shows Teslas with FSD are only about three times safer, and the methodology still omits factors such as vehicle age and short‑window crash reporting.

Interviews with nine former Tesla data labelers – the workers who review raw video to train the AI – revealed that seven would not ride in a vehicle operating on FSD. They reported frequent instances of the system exceeding speed limits and failing on routine road conditions, issues they said were deprioritised by engineers. The labelers’ testimony contrasts sharply with Tesla’s public claims and comes as the company expands FSD to new markets, including China, while the system remains a Level‑2 driver‑assist technology requiring constant supervision. The findings raise questions about the reliability of Tesla’s marketing and could prompt closer regulatory scrutiny.