Comment by RandomLensman

Reward hacking/going for unanticipated solutions is nothing new in ML/AI, already much simpler systems have done/do "weird" things (gut feeling is that iterative and ensemble use majes the surface for that much larger).