Abstract: Artificial intelligence evidence currently enters litigation primarily through conversion into existing evidence forms or as a distinct evidentiary category. However, its admissibility review often emphasizes relevance and objectivity while neglecting the legality dimension, compounded by the absence of specific review rules. Establishing a framework for legality review rules necessitates development across three dimensions, including the determination of categories, the object of review and the review procedure. The specific scope of AI evidence must be clearly delineated by optimizing classification and naming mechanisms, abandoning the current inconsistent terminology in favor of a unified concept of AI evidence with sub-categorization based on technical generation logic and functional evidentiary attributes. The legality of evidence must be redefined within the hierarchical structure of evidence attributes. On this basis, rules for reviewing element attributes and guidelines for evidence adjudication should be established to maintain a balance between the pursuit of truth and the protection of human rights. The legality review aspect of the norms on the basis for the final decision covers the content of review and review procedures, which is the core dimension of the legality review rules for artificial intelligence evidence. Judicial authorities should review the source, process, and outcome of AI evidence and apply mandatory or discretionary exclusionary effects as appropriate to promote the standardized application of AI evidence.