Consider how widely affected contemporary machine learning models are by adversarial examples [1]. I don't know the specific approach used by Tesla or their rel

Consider how widely affected contemporary machine learning models are by adversarial examples [1]. I don't know the specific approach used by Tesla or their release process, but I would not be surprised at all if their software has similar shortcomings. Granted, a ton of active research is going on trying to prevent these sorts of problems, but it definitely isn't a solved problem. Basically, ask yourself this: how robust is this software? How are you measuring that? [1] https://arxiv.org/pdf/1712.07107.pdf https://arxiv.org/pdf/1712.07107.pdf