I've been on plenty of machine learning projects and have seen the boss make promises that the engineers are expected to keep. This is exactly what it looks like. The outcome is always overfitting a simulation or data set somewhere so that it looks like it's working, but the real world performance ends up being worse. Unless the engineers get stubborn and declare themselves to have failed (which can actually happen).

But in the projects I did, it was some dumb sales recommendation system. These people are doing it with self driving cars.