Abstract: Artificial intelligence (AI) is a key technology for advancing the modernization of the nuclear power safety supervision system. To achieve the modernization of nuclear power safety supervision, it is necessary to systematically analyze the mechanism of AI's impact on the nuclear power safety supervision system and construct a modern supervision logic adapted to the characteristics of the nuclear power industry. The study finds that the mechanisms by which AI empowers the modernization of the nuclear power safety supervision system are mainly reflected in the synergy of supervision subjects, the scientization of supervision means, the legalization of supervision order, and the digital intelligence of the supervision process. However, the application of AI faces four major challenges: the institutional paradox between data sharing and information silos, the practical conflict between algorithmic black boxes and transparent supervision, the development gap between technological iteration and institutional rigidity, and the ethical dilemma between data-driven approaches and ambiguous responsibilities. Therefore, to address the above issues, we should construct a "three-dimensional collaborative" governance system for nuclear safety supervision data, innovate the "penetrating" supervision paradigm for nuclear safety, establish a "dynamically adaptive" institutional governance mechanism for nuclear safety, and build a "responsibility-anchored" ethical governance framework. This provides a decision-making reference for global nuclear power safety supervision institutions in their digital transformation.