Abstract: While the judicial application of generative artificial intelligence has significantly improved adjudication efficiency and promoted consistent judgments for similar cases, its inherent “black box” characteristics and data bias pose systematic challenges to judicial justice and openness, the dominant status of judges, and judicial credibility. The fundamental crux lies in the potential erosion of judicial value rationality by technical rationality, and the structural dissolution of traditional procedural legitimacy by algorithmic decision-making. Therefore, it is imperative to construct a dynamic regulatory framework with risk and function as dual coordinates. This can be achieved by integrating a liability attribution system of “risk stratification and composite subjects”, industrial self-regulatory norms of “ethical preposition and bottom-line guarding”, and collaborative governance tools of “sandbox regulation and dual-track transparency”, so as to ensure that technological empowerment always serves the humanistic core of the judiciary and realize the prudent and balanced development of digital rule of law.