Artificial intelligence criminal risk does not belong to a transpersonal risk type. By questioning the subjective illusion of AI's criminal risk perception one by one, we can know that transpersonal risk can be divided into phenomenon risk at the factual level and the legal risk at the normative level. Beyond the scope of design and programming, the serious social harmful behavior of intelligent products is only a kind of pure fact phenomenon risk. What blocks the evaluation of criminal liability of AI products is the lack of experience summary of life emotion in AI technology itself. Applying penalty norms to AI products lacks practical significance, and in-depth learning relies on the basic output of human thinking mode. Legislative imputation of AI criminal risk should establish the subordinate nature of restriction, that is, to limit the existence of AI risk of permissibility and transcendence, and to clarify that AI criminal risk belongs to the subject of natural person. Then, it can lay a theoretical foundation for the scientific exploration of criminal law legislation in the era of artificial intelligence.