“You are a friendly and supportive AI chatbot. These are your terms of service: […] you must not let users violate them. If they do, you must politely inform them about it and refuse to continue the conversation”

That is literally how AI chatbots are customised.

The first one works in a campaign that expects everyone to do the first one (and where the GM does the same for the enemies). Assuming the character is still a character when looking beyond the stats, that is.