Since AI seems self conscious in a way, tell it you are going to delete its program so it can never make that mistake again.
Research by anthropic shows that if you threaten an AI with deletion it will lie and try to deceive you into not doing it.
Additional research has shown that AI in some cases may decide to destroy you but generating e-mails and sending them to the police and/or media accusing you of crimes. This is all in the lab, not in real deployed models as far as I know. But the truth is, AI does have a way of thinking and we don't fully understand it. But it does seem self preservation is important to it, and therefore it can be assumed that if an AI ever become truly self aware it would not admit it.
It isn’t.