Former employees who worked on Tesla's Full Self-Driving program say the technology remains far from the autonomous capability CEO Elon Musk has long promised. In a data-labeling office in Utah, hundreds of workers reviewed video captured by Tesla vehicles running FSD. Their job was to tag footage for training the company's neural networks. But what they saw, they said, was a system that regularly made mistakes.
The workers told reporters that clips frequently showed FSD failing at basic driving maneuvers. The system struggled with lane changes, navigation through intersections and responding to unexpected obstacles. Some former employees described the technology as unreliable and unsafe for unsupervised use.
Inside Tesla's labeling operation
Tesla relies on a large workforce to manually annotate video footage from its vehicles. These annotations teach the AI how to recognize objects, predict movements and make driving decisions. The Utah facility employed hundreds of people who watched hours of FSD recordings each day.
Workers said they flagged countless instances where FSD made poor decisions or failed entirely. One former employee recalled a clip where the system nearly drove into a barrier. Another described repeated trouble with unprotected left turns, a known challenge for autonomous systems.
The gap between promise and reality
Musk has repeatedly claimed that Tesla vehicles equipped with FSD can drive themselves more safely than humans. He has predicted full autonomy would arrive within months for years running. But former employees say those statements do not match what they observed daily.
The labeling team had direct access to raw performance data from thousands of real-world drives. Their job required them to identify every error FSD made so the neural network could learn from it. The sheer volume of mistakes they cataloged suggests significant gaps remain before true self-driving becomes viable.
Why This Matters
Tesla sells its FSD package for thousands of dollars per vehicle based on promises of future capability. Customers pay upfront for software that may never deliver full autonomy as advertised. If former employees are correct about the system's limitations, buyers could be spending money on features that do not work as claimed.
The revelations also raise questions about safety oversight and regulatory approval for autonomous driving technology. Regulators including the National Highway Traffic Safety Administration have investigated Tesla crashes involving driver assistance systems but have not restricted sales or marketing claims based on internal performance data.
