Abstract: With the improvement of algorithms, arithmetic power and big data technology, DeepSeek-like artificial intelligence technology has developed rapidly and iteratively. Currently, DeepSeek-like artificial intelligence has triggered the copyright infringement of model training while enhancing economic and social benefits. Among them, in the technical operation layer, the logical structure of “collection-training” leads to the improper use of works training; in the infringement determination layer, the application of the infringement determination standard of “contact+substantial similarity” fails to work; in the defense layer, the fair use of copyright fails to apply. In the face of the copyright infringement risk of DeepSeek-like artificial intelligence model training, the principle of fault liability should be adopted to recognize copyright infringement liability in order to promote the innovative development of AI. At the same time, the copyright infringement determination standard should be optimized, new types of fair use of data mining should be introduced, “new safe harbor rule” to encourage commercial model training to appropriately allocate the responsibility of service providers should be set up, supplemented by copyright infringement liability insurance and compensation fund system to disperse the risk of copyright infringement, so as to achieve the goal of copyright infringement regulation and promote the development of DeepSeek-like artificial intelligence.