Artificial Intelligence (AI)-assisted inventions pose significant challenges to patent novelty examination, primarily manifested in three dimensions: the scope of prior art, the principle of individual comparison, and examination criteria. Firstly, the application of AI has triggered an exponential growth in prior art volume, potentially distorting novelty examination into a tool for patent competition. Secondly, the principle of individual comparison reveals limitations when addressing cross-domain technological convergence, complicating the identification of innovative features in AI-assisted inventions. Thirdly, AI-generated technical documentation increases the complexity of examination processes. To address these challenges, this study proposes reconstructing the identification criteria for prior art through the introduction of a “substantive accessibility” standard. Concurrently, it advocates establishing a cross-domain comparison principle to clarify technical relevance assessments and predictability tests for prior art combinations. Furthermore, the principle of technical substantive equivalence and the integration of intelligent examination tools are recommended to strengthen disclosure obligations. These measures aim to ensure that AI technology fosters innovation while upholding the fairness and efficacy of the patent system.