I would say Tesla's "FSD". For years they have promised it and cashed in the money, making it even non-transferable. FSD is still years away from working.
Part of the problem with FSD is that it is difficult to make and test incremental improvements on a larger scale. At some point you need to send a big fleet of cars and let an "average" driver drive it without any special safety measures
Another part of the problem is public perception of safety, people are willing to accept fatalities due to human mistakes but even minor accidents caused by FSD are highly publicized
The other side of this is that a machine may prevent other accidents that a human cannot prevent. As long as our machines aren't truly superhuman we will have to find an optimal middle-ground (between human and machine control).