二十一世纪当下，人工智能技术取得了新的突破，生成式人工智能应运而生。相比传统人工智能技术，其在模仿人类创造和生成新内容等方面具备前所未有的卓越性能，具备出色的语义理解、自然语言生成与问题解答能力，已逐渐应用于人类工作和生活的方方面面，而现行刑事法律对于生成式人工智能相关犯罪的规制仍大多以网络犯罪为参照，学界对于有关的犯罪类型、风险判定也无统一观点。因此，随着生成式人工智能应用场域的迅速扩展，需要从学理上探讨其给刑事法律带来的新的挑战与风险，为现有刑法规制体系的完善提供智力支持。

本文在对生成式人工智能的相关概念、发展历程以及运行原理进行梳理的基础上，依据现行刑法的规制能力将涉生成式人工智能犯罪分为三种类型，一是“工具使用型”犯罪，即可以完全依靠现行刑事法律规制的类型；二是“强大助力型”犯罪，即现行法律无法完全规制的类型；三是“人工智能失控型”犯罪，即现行法律根本无法规制的类型。对于生成式人工智能所带来的刑事法律风险，以其运行流程为依据，按照其内容生成的运行顺序，将其分为内容生成前的数据不当采集处理风险、内容生成中的生产过程的不可控风险和内容生成后的生成技术滥用风险。内容生成前，算法设计和数据训练阶段均存在潜在的数据风险；内容生成中，不可控的生成过程可能导致价值偏向、事实虚构和侵犯知识产权等风险；内容生成后，滥用行为可能引发制造犯罪工具、传授犯罪方法和更新犯罪手段等风险。最后，针对生成式人工智能的相关犯罪及其产生的风险，提出相应的刑事治理路径。一方面要明确规制思路：在立法与解释之间寻求平衡，既要确保刑事法律的稳定性，又要适应科技的快速发展；在风险治理与科技发展之间寻求平衡，既要有效防控风险，又要避免过度干预科技创新；在前瞻性与谦抑性之间寻求平衡，既要预见未来可能出现的风险，又要保持刑法的谦抑性，避免对科技创新的过度打压。另一方面需厘清主体责任：在生成式人工智能运行全链条设置刑事合规机制，对内容生成前端、中端和后端的主体责任进行科学界定，在研发、运营、使用全过程明确各主体的刑事责任。对于研发者，应明确其在算法设计和数据处理中的法律责任；对于运营者，应规范其在内容生成和发布过程中的行为；对于使用者，应明确其在使用生成内容时的合法性和合规性。

In the twenty-first century, Artificial Intelligence technology has made new breakthroughs, and Generative Artificial Intelligence has emerged. Compared with traditional AI technology, it has unprecedented performance in imitating human creation and generating new content, has excellent semantic understanding, natural language generation and question answering ability, and can meet the needs of human beings in various scenarios, and has been gradually applied to all aspects of human work and life. The current criminal law for generative AI-related crime regulation is still mostly based on cybercrime as a reference, and there is no unified viewpoint in the academic community on the type of crime and risk determination. Therefore, with the rapid expansion of the application field of generative artificial intelligence, it is necessary to explore the new challenges and risks it brings to the criminal law from a theoretical perspective, so as to provide intellectual support for the improvement of the existing criminal law system.

Based on the concepts, development history and operation principles of generative AI, the article divides generative AI crimes into three types according to the regulatory capacity of the current criminal law: First, "tool-use" crimes,types that can be fully regulated by existing criminal laws; second, "powerful assistance" crimes, types that cannot be fully regulated by existing laws; and third, "AI out-of-control" crimes, types that cannot be regulated by existing laws at all. With regard to the criminal legal risks brought about by generative artificial intelligence, based on its operational process, according to the operational order of its content generation, it is divided into the risk of improper data collection and processing before content generation, the uncontrollable risk of the production process during content generation, and the risk of generative technology abuse after content generation. Before content generation, there are potential data risks in both the algorithm design and data training phases; during content generation, the uncontrollable production process may lead to risks such as value bias, factual fictions, and infringement of intellectual property rights; and after content generation, abusive behaviors may trigger the risks of manufacturing criminal tools, teaching criminal methods, and updating criminal means. Finally, the corresponding criminal governance path is proposed for the crimes related to generative AI and the risks it generates. On the one hand, it is necessary to clarify the idea of regulation and seek a balance between legislation and interpretation, not only to ensure the stability of criminal law, but also to adapt to the rapid development of science and technology; to seek a balance between risk management and scientific and technological development, not only to effectively prevent and control risks, but also to avoid excessive intervention in scientific and technological innovation; to seek a balance between foresight and modesty, not only to foresee the risks that may arise in the future, but also to maintain the modesty of criminal law to avoid the excessive suppression of scientific and technological innovation. Seek a balance between foresight and modesty, not only to foresee the possible risks in the future, but also to maintain the modesty of the criminal law, to avoid excessive suppression of science and technology innovation. On the other hand, it is necessary to clarify the responsibility of the main body, set up a criminal compliance mechanism in the whole chain of generative AI operation, scientifically define the responsibility of the main body in the front-end, middle-end and back-end of content generation, and clarify the criminal responsibility of the main body in the whole process of research and development, operation, and use. For  developers, their legal responsibilities in algorithm design and data processing should be clarified; for operators, their behaviors in the process of content generation and distribution should be regulated; and for users, their legality and compliance in the use of generated content should be clarified.