Diagnostic AI has undergone a technological leap from being “rule-driven” to “data-driven”. With its inherent features—autonomy, an algorithmic black-box, and dynamic evolution—it poses structural challenges to the traditional medical malpractice liability system centered on the “prevailing medical standard”. To address these challenges, the use of diagnostic AI should be positioned as the adoption of a “new therapy” to establish its legal basis. It is necessary to clarify that its outputs do not directly equate to the “prevailing medical standard”, and that the connotation of this standard will also evolve. On this basis, we propose a dynamic fault-determination framework centered on the “duty of prudent re-evaluation” . This framework shifts the focus of assessment to an evaluation of the process by which medical professionals fulfill their obligations of prudent use and substantive verification of algorithmic outputs, thereby responding to the needs of liability determination in the current “human-computer collaboration” model.