That model will only be able to detect prompts it is trained on which is the security strategy they already use and doesn’t work