I think you're interpreting this work too narrowly. Don't think of it as having found the one thing that can go wrong with autopilot. Think of it as one exampl

I think you're interpreting this work too narrowly. Don't think of it as having found the one thing that can go wrong with autopilot. Think of it as one example piece of evidence for how (non) robust the autopilot is.