An attack, typed directly by the user, that tries to talk a model out of its own trained safety behavior for one request.
Continue to AI University →