Even Tesla’s Own AI Trainers Don’t Trust Its Self-Driving Tech – And For Good Reason Too

Image Courtesy: Getty Images

Tesla’s Full Self-Driving (FSD) software has been promoted by the company as being significantly safer than human drivers, but former employees involved in training the system say the technology still struggles with basic driving tasks and is far from delivering fully autonomous vehicles at scale.

Hundreds of Tesla data labelers review footage collected by vehicles using FSD, identifying both successful and problematic driving behavior to help train the company’s AI systems. Interviews with former employees and a former Tesla self-driving engineer revealed recurring issues, including failures to stop for school buses, difficulties recognizing emergency vehicles, speeding incidents, and near-misses involving pedestrians. Their accounts challenge public claims made by Tesla executives about the technology’s readiness and safety, according to a special report by Reuters.

The report also raises questions about Tesla’s frequently cited safety statistics. Tesla executives, including CEO Elon Musk, have claimed FSD is up to ten times safer than human drivers. However, Reuters found that the company’s methodology relies on comparisons that several traffic-safety researchers described as misleading.

One key issue involves Tesla comparing crashes that triggered airbag deployments in FSD-equipped vehicles against a broader federal dataset that includes less severe accidents. Researchers who reviewed the methodology said a more direct comparison would reduce Tesla’s claimed safety advantage significantly. Experts also noted that Tesla compares its relatively new vehicle fleet to the overall U.S. vehicle population, which is much older and generally less equipped with modern safety technologies.

Former employees also described extensive efforts behind Tesla’s public robotaxi demonstrations. Ahead of major launches, workers reportedly spent months mapping routes, labeling road features, and training the software on specific local hazards. The process contrasts with Musk’s long-standing claim that Tesla’s AI can operate anywhere without relying on detailed local mapping.

Several former staff members said they would not personally trust FSD to drive them. Some recalled reviewing videos showing Teslas striking animals, speeding well above posted limits, or requiring human intervention to avoid collisions. One former engineer who reviewed crash data described Tesla’s public safety claims as unreliable.

Tesla continues to state that FSD requires active driver supervision and is not currently autonomous. Meanwhile, federal regulators maintain multiple active investigations into crashes involving Tesla’s driver-assistance systems.

Despite years of promises that fully autonomous driving is imminent, former employees and independent safety experts told Reuters that significant technical and safety challenges remain before Tesla can deliver self-driving vehicles on a large scale.