Abstract: Faced with the criminal risks associated with large generative AI models, the existing regulatory schemes based on interpretative theory and legislative theory may suffer from uncertainties in the application of criminal charges and difficulties in determining the object of the behavior. The foundation and essence of the development of large models lie in data, and the nature of their criminal risks is data risk. Therefore, the data crime model should be applied to regulate these risks. This model is reasonable based on the theory of the right tree and operable based on data law. Hence, the criminal risks of large models can be regulated by reshaping data crimes. In terms of interpretative theory, three major methodologies should be applied: advocating for a dual-layer legal interest in data management order to achieve independent regulation of data crimes; preventing and controlling different data processing behaviors at the input, operation, and output stages of large models to achieve full lifecycle regulation of data crimes; and clarifying unacceptable, significant, and general risks of large model data to achieve classified and graded regulation of data crimes. In terms of legislative theory, by establishing the crime of illegally processing data with attributes of statutory offense, consequential offense, and tiered punishment, the drawbacks of existing data crimes can be thoroughly addressed, achieving precise management of data risks associated with large models.