A recent and comprehensive investigation by the Reuters news agency, brought to light by Electrek, has shaken the very foundations of Tesla’s ‘Full Self-Driving’ (FSD) system – particularly its much-vaunted safety statistics. The report reveals a deeply flawed methodology underpinning the data. But what is even more worrying is that even Tesla’s own data labellers – the employees who train the AI system and know its limitations best – do not trust the technology to drive itself.
The Reuters exposé: Flawed statistics and internal mistrust
The Reuters investigation is based on interviews with current and former Tesla employees, internal documents and data analyses. It paints a picture that stands in stark contrast to the public statements made by Tesla and Elon Musk, who portray FSD as significantly safer than human drivers. The shortcomings in data aggregation and interpretation could result in a misleading representation of FSD’s actual safety performance. This is particularly problematic as safety is a crucial factor for the acceptance and approval of autonomous driving systems worldwide, and particularly in the regulated European market.
The scepticism of the AI trainers is an alarming sign. These specialists spend hours every day annotating the most complex and challenging traffic scenarios to train the neural network. They see the so-called ‘edge cases’ – the borderline situations where FSD struggles or makes mistakes. When those working closest to the technology are reluctant to use it themselves, it speaks volumes about the system’s current maturity and reliability.
Implications for the European FSD rollout
For Europe, the implications of this revelation are serious. Whilst Tesla FSD (Supervised) is being rolled out gradually in some European countries such as Belgium, the Netherlands and Lithuania (see Breakthrough in Belgium), regulatory bodies such as the UNECE (United Nations Economic Commission for Europe) are paying particularly close attention to safety evidence. Inaccuracies or a lack of transparency in safety statistics could significantly slow down or even halt the approval process. Scepticism among the authorities themselves could reinforce the arguments against a swift and far-reaching approval of FSD in Germany and other EU countries.
In the past, there have already been complaints and concerns regarding Tesla’s FSD promises. For instance, Tesla was sued in China for ‘Full Self-Driving’ fraud (see FSD promises under fire) and an Oracle manager won a US lawsuit against Tesla over FSD promises (see [US lawsuit against Tesla](/artic