ChatGPT-like system refers to the content-generating artificial intelligence (AIGC) represented by ChatGPT, which combines generative pre-training with algorithmic language conversion. ChatGPT-like system not only promotes the development of human society, but also brings criminal risks. When ChatGPT-like systems obtain corresponding information to generate certain content, it is easy to break away from human control and intervention and improperly infringe on the confidentiality of other data. At the same time, if the content generated by ChatGPT-like system contains illegal and criminal information, or users improperly use the information generated by ChatGPT-like system, it may also involve corresponding criminal risks. However, ChatGPT-like system does not have the qualification of the subject of criminal responsibility, that is, the subject that may ultimately bear criminal responsibility is not ChatGPT-like system, but the developer or user related to it. In the development and use of ChatGPT-like system, to prevent technical risks and make the technology "good", the developers should fulfill the obligations as guarantors, foresee the possible harmful results caused by ChatGPT-like system as much as possible, and take corresponding countermeasures to prevent them. Users should not intentionally induce ChatGPT-like system to break through the ethical and legal bottom line and output illegal and criminal information. If the developer fails to foresee and prevent ChatGPT-like system from outputting illegal and criminal information to the greatest extent, or the user intentionally induces ChatGPT-like system to output illegal and criminal information, he shall bear corresponding criminal responsibility. From the macro level, when regulating criminal risks involved in ChatGPT-like system, the criminal law principle of modesty should be observed to avoid hindering technological innovation and development. From the micro level, in view of the content and characteristics of criminal risks in ChatGPT-like systems, it is necessary to construct a hierarchical governance mechanism: develop and implement compliance plans through ChatGPT-like system research and development enterprises, and explore preventive governance strategies for the criminal risks; strengthen administrative supervision, and integrate scientific and technological supervision means into administrative supervision system, to reduce the possibility of criminal risks; the use of criminal governance measures must strictly adhere to the precondition that prior legal and regulatory provisions are ineffective, the assessment of criminal illegality must be conducted within the framework of the principle of legality in criminal law, and administrative illegality should not be used to replace criminal illegality.