The subject of generative artificial intelligence is built on relatively sophisticated neural network technology. Driven by the deep learning of pre-trained data and human feedback data, it exhibits unprecedented human-like attributes and inexplicability in both its generation process and terminal manifestations. From the perspective of criminal law, the risks posed by generative artificial intelligence can be subdivided into two dimensions: endogenous criminal risks created by generative artificial intelligence subject itself in the absence of external behavioral intervention, and exogenous criminal risks induced by external factors. Criminal law intervention in both the endogenous and exogenous risks of generative artificial intelligence demands a prudent response, and the value orientation of criminal law should be established on the basis of weighing the values of balancing development and security. An absolute bias toward either the positive criminal law view or the negative criminal law view is inherently limited. Instead, we should adhere to the more appropriate adaptive criminal law view, that is, on the premise of upholding the position of criminal law as a guarantee law, we oppose hasty legislative regulation of the criminal risks induced by generative artificial intelligence to avoid curbing the development dividends of the digital economy era due to excessive expansion of the scope of crimes. Meanwhile, we must keep a close watch on the iterative evolution of generative artificial intelligence and actively adjust traditional criminal law theories to address the potential negative harms it may cause. In the debate between instrumentalism and subjectivism, it is necessary to clarify the inaccuracy of pure instrumentalism and pure subjectivism in defining the attributes of generative artificial intelligence. At the current stage, we should neither overstate its degree of autonomy by treating it as an independent subject of criminal liability, nor adhere to traditional views rigidly by regarding it merely as a passive tool. Under the correct guidance of the adaptive criminal law view, facing the inadequacy of the traditional criminal law causality theory in responding to generative artificial intelligence, we should conduct an in-depth analysis of the causal relationships arising from interactive participation, emphasize the pivotal position of the causative force theory in the traditional criminal law causality theory, and reveal the specific manifestations of causative force relationships in the process of criminal law imputation amid the impact of criminal risks posed by generative artificial intelligence. On this basis, by further expanding the scope of subjects bearing causative force, incorporating behavioral correlation into the scope of causative force judgment, and distinguishing the degrees and types of causative force, we should conduct an adaptive and objective interpretation of the causal relationships between the acts of relevant subjects involved in criminal risks of generative artificial intelligence, the operation of generative artificial intelligence, and the harmful consequences thereby caused, through the innovation and reconstruction of criminal law theories in the digital and intelligent era. This lays a theoretical foundation for the dynamic criminal law regulation of generative artificial intelligence and properly addresses the challenges posed by generative artificial intelligence to criminal law imputation.