After the large-scale application of generative artificial intelligence technology, a large amount of information generated by it enters criminal investigation and other scenes, becoming an important material for case adjudication. This kind of information has the characteristics of black box algorithm, fuzzy subject and creative content, which conflicts with the traditional evidence rules with human dominant generation, objective fact record and fixed carrier storage as elements, thus bringing difficulties such as evidence ability examination, proof evaluation and evidence identification. At the level of evidence ability, authenticity examination encounters the difficult problems of invalid source qualification and failure of content verification; relevance identification falls into the deadlock of substantial association judgment due to difficulty in traceability; legality examination faces the problems of vague subject responsibility, insufficient procedure adaptation and coexistence of content compliance risks. At the level of probative power, the lack of judgment standard and vague hierarchical positioning lead to great risk of different judgments in the same case. At the practical level of evidence, it is difficult to guarantee the integrity of evidence preservation, and the authentication procedure lacks special rules to ensure its reliability. In order to solve the dilemma, we should start from the three principles of evidence, link up the characteristics of generative technology, and take the traceability of technology as the anchor point to construct the analysis framework of underlying logic, method logic and value logic. At the practical level, based on the generation mode and evidence use, the classification is carried out out: the auxiliary generation type is distinguished from the independent generation type, the fact proof type and the procedure auxiliary type, the rules such as the three-nature examination of evidence ability, the differential evaluation of proof power, the adaptation of practical operation process, etc. are constructed respectively, and the distribution of burden of proof, the standardization of cross-examination procedures, the technical path of integrity preservation and the expert auxiliary mechanism are clarified.