Comment by ramoz

One concern nobody likes to talk about is that this might not be a problem that is solvable even with more sophisticated intelligence - at least not through a self-contained capability. Arguably, the risk grows as the AI gets better.