When I think of self-driving car technology, I think of all of these inputs (maps, cameras, GPS, etc.) being fed into a world model from which inferences are drawn.[1] Here, the car could have known from cameras and maps that there was a lane split, and should have been able to detect that the line following was giving a conflicting inference compared to those other sources. The question is: does Tesla actually do that? Because if not, it's not really self-driving car technology. It's not a glitch in a product that's 90% of the way there; it's a fundamentally more basic technology.
[1] See https://www.ri.cmu.edu/pub_files/2009/6/aimag2009_urmson.pdf at 21-23.