Apparently profanity, factual intimidation and name-calling work when AI tries to scam you.