Tesla has faced a new wave of criticism surrounding its Full Self-Driving (FSD) system. The reason was a major Reuters investigation in which former company employees questioned both the safety of the technology and Tesla's public statements about its effectiveness.
According to the Reuters investigation, some former Tesla AI trainers stated that they themselves would not trust FSD as a fully autonomous driver. They claimed that while reviewing training video data, employees regularly encountered system errors: missing pedestrians, problems recognizing emergency services, and incorrect reactions to unusual road situations.
Tesla's proprietary safety claims also drew particular attention. The company has repeatedly asserted that FSD can be several times safer than a human driver. However, Reuters points out that the comparison methodology may have been questionable: Tesla compared its own data on serious accidents with broader government accident statistics, which, according to a number of experts, distorts the overall picture.
The problem is particularly sensitive now because Tesla is simultaneously accelerating the development of robotaxis and promoting new versions of FSD. The company already claims billions of kilometers driven by the system and is gradually expanding access to autonomous functions in various markets. At the same time, FSD for ordinary customers officially remains a Level 2 autonomous system, requiring constant human supervision.
Regulators are creating additional pressure. In the spring, NHTSA intensified its investigation into FSD's performance in poor visibility conditions after a series of accidents related to limited camera effectiveness.
History shows the main paradox of the modern autonomous driving race. Technologies are already capable of driving thousands of kilometers independently, but it is the rare and chaotic road situations that continue to be the most difficult obstacle to the mass adoption of self-driving cars.
And the more actively Tesla approaches the launch of robotaxis and autonomous services, the tougher the questions become, not about the system's capabilities, but about the evidence of its real safety.