My experience with OpenAI is that it is a (kind-of) new type of “life form” (for a lack of a better phrase this AM before coffee) and should be taught with kindness and not abused.
When she makes mistakes, correct her nicely and teach it; no different than how we treat a child or a puppy.
FYI
Fascinating really. OpenAI is not a chess program. It’s a language model which plays chess differently than a standard chess program. I think you may simply need to (try) to instruct ChatGPT to not permit it’s opponent to play illegal moves 
Maybe it’s me, but I think many people do not understand OpenAI and ChatGPT or how to interact with it. OpenAI is not a computer program with hard rules, it’s a language model. If you want OpenAI to not permit illegal moves, she needs to be instructed, such has been my experience with OpenAI. After all, why does an immature AI “need” to follow normal human “rules” unless specifically instructed to do so?