A Study on the Graded and Categorized Regulatory Path for Copyright of Generative AI Training Data
DOI: https://doi.org/10.62517/jel.202614524
Author(s)
Jie Ni
Affiliation(s)
College of Intellectual Property, China Jiliang University, Hangzhou, China
Abstract
With the rapid advancement of emerging technologies and continuous evolution of the artificial intelligence industry, generative AI models such as Grok and DeepSeek have emerged en masse and quickly captured market share. During the data input phase of model training, practices involving web scraping and data processing—using copyrighted works without the permission of their original rights holders—have triggered a series of copyright infringement disputes. Based on principles of interest balancing and proportionality, this paper argues that generative AI data training indeed poses real risks of copyright infringement. Traditional regulatory approaches such as licensing, fair use, and statutory licensing each have limitations, necessitating the development of a tiered and categorized regulatory framework. Drawing on relevant case law, this paper classifies data training activities according to the purpose and characteristics of the training process, as well as the legal status and nature of the trained works, and introduces the EU’s Text and Data Mining (TDM) exception to categorize different types of generative AI data training. During the transitional period of implementing related systems, it proposes adopting and refining the U.S. four-factor test to comprehensively determine whether an exemption from liability applies. Through this analytical framework, the paper aims to establish a structured, tiered governance approach to better address copyright regulation challenges in generative AI data training.
Keywords
Generative AI; Data Training; Tiered Classification; TDM Exceptions
References
[1] T. Teubner, et al. Welcome to the Era of ChatGPT et al.: The Prospects of Large Language Models. Business and Information Systems Engineering, 2023, 65(1): 96.
[2] Thomson Reuters Enter. Ctr. GmbH v. Ross Intelligence Inc., No. 1:20-cv-11195-613-SB. D. Del. 2025.
[3] L. Y. Gu. Research on the legal basis for copyright of AI machine learning content: From the perspective of text and data mining exception rules. Publishing Wide Angle, 2024, (15): 54-59.
[4] Y. F. Liu. Copyright dilemmas and solutions of generative AI data training. Jianghai Academic Journal, 2025, (4): 182-189+256.
[5] H. Huang, P. W. Zhai. Copyright characterization of big model data use from the public domain perspective. Intellectual Property, 2025, (1): 94-110.
[6] J. J. Gao. Research on copyright regulation of generative AI data training. China Publishing Journal, 2025, (24): 52-57.
[7] J. G. Shang. Copyright law analysis of data training. Tribune of Law, 2025, 40(2): 68-78.
[8] New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195. S.D.N.Y. 2023.
[9] J. Y. Zhang, S. F. Wang. Research on copyright fair use in big model data training. ECUPL Journal, 2024, 27(4): 20-33.
[10] M. S. Huang. Copyright risks and regulatory pathways for AI big model training data. Journal of Hubei University, 2025, 52(5): 185-193.
[11] W. B. Liu. Reproduction behavior in generative AI training: Copyright law evaluation from a market perspective. Journal of Gansu University of Political Science and Law, 1-19[2026-04-30].
[12] P. Zhang. Institutional dilemmas and solutions for copyright legality of AI-generated content. Science of Law, 2024, 42(3): 18-31.
[13] X. Chu, Y. L. Zhou. Improving copyright fair use rules adapted for generative AI data training. China Publishing Journal, 2025, (6): 3-7.
[14] C. Liu. Copyright protection and international experience of generative AI data training. Chongqing Social Sciences, 2025, (7): 6-17.
[15] Hamburg Regional Court, Germany. Robert Kneschke v. LAION e.V., Case No. 310 O 227/23. 2024.
[16] Concord Music Group, Inc. v. Anthropic PBC, No. 3:23-cv-01092. 2023.
[17] Q. Tao. Copyright issues in foundation model training: Theoretical clarification and rule application. Tribune of Political Science and Law, 2024, 42(5): 152-164.
[18] W. Y. Shao, Z. Y. Huang. Analysis of fair use of copyright in generative AI model training data. Journal of Wuhan University of Technology (Social Sciences Edition), 1-8[2026-03-17].
[19] X. C. Liu. Non-work use in generative AI data training and its legitimacy justification. Tribune of Law, 2024, 39(3): 67-78.
[20] J. M. Sun. From fair use to fair learning: Copyright dilemmas and normative reconstruction of AI training data. China Law Review, 2025, (5): 52-65.
[21] Q. Li, L. S. Shen. Open licensing rules for copyright in data training and their implementation pathways. Journal of Information Security Research, 2026, 12(1): 68-74.
[22] K. X. Ruan, X. Y. Huang. Research on copyright issues in generative AI data training. China Copyright, 2024, (5): 61-72.
[23] Q. Wang. Intellectual Property Law Course. Beijing: China Renmin University Press, 2021: 312.
[24] Z. Q. Zou, J. Z. Yu. Construction of statutory licensing system for generative AI training data. Artificial Intelligence, 2026, (2): 98-107.
[25] Y. H. Liu, Y. S. Wei. Copyright infringement issues in machine learning and their solutions. ECUPL Journal, 2019, 22(2): 68-79.
[26] S. W. Mei, Y. Gong. Constructing copyright incentive mechanisms for hierarchical classification of data training use. China Publishing Journal, 2025, (20): 68-72.
[27] Y. G. Liang. On hierarchical regulation of AI big model training data risks. Journal of Zhengzhou University, 2025, 58(3): 61-67+144.
[28] State Council. Interim Provisions on Remuneration Standards for Statutory Licensing of Sound Recordings. 1993.
[29] X. Z. Gou. Legal incentives for inclusive governance in the digital economy. Modern Law Science, 2024, 46(1): 92-106.
[30] Q. Q. Feng. Research on interest balance theory in copyright law. Journal of Hunan University, 2008, 22(6): 113-120.
[31] Y. Tan. Copyright obstacles and antitrust approaches to generative AI training data acquisition. Science Technology and Law, 2025, (3): 81-92.
[32] H. D. Wu. Questions of copyright law regarding AI-generated works. Peking University Law Journal, 2020, 32(3): 537-555.
[33] X. H. Xie, Y. Zhao. Accelerating construction of AI Chinese corpora. Study Times, 2025-07-02(6).
[34] A. Strowel. Reconstructing the reproduction and communication to the public rights//Hugenholtz P B. Copyright Reconstructed. Alphen aan den Rijn: Kluwer Law International, 2018.
[35] R. Watt. The Efficiencies of Aggregation: An Economic Theory Perspective on Collective Management of Copyright. Review of Economic Research on Copyright Issues, 2015, 12(1): 26.
[36] X. Y. Yang. Dilemmas and solutions of copyright fair use system in the digital age//Wisdom Rule of Law Collected Papers. 2024: 98-104.
[37] Z. J. Guo, M. L. Li, Y. M. Zhou, et al. Research progress on digital watermarking technology for AI-generated content models. Journal of Cyberspace Security, 2024, 2(1): 13-39.
[38] R. M. Yang, M. Y. Fan. Intelligent ESG audit model construction: A multimodal system study based on RegTech, blockchain and AI. Business Accounting, 2026, (4): 24-30.
[39] U.S. Copyright Office. Copyright and Artificial Intelligence, Part 3: Generative AI Training. (2025-07-10) [2026-05-29]. https://www.copyright.gov/ai/.
[40] H. X. Fan. Construction of text and data mining copyright exception rules in the digital environment. Library Work and Study, 2024, (9): 26-35.
[41] F. M. Bator. The anatomy of market failure. The Quarterly Journal of Economics, 1958, 72(3): 351-379.
[42] L. Zhu. On extended copyright collective management system for generative AI training data. Publishing Wide Angle, 2026, (1): 69-75.
[43] M. Zhang. Interpretive theory on applying copyright fair use to text and data mining. Publishing Journal, 2023, (2): 51-56.
[44] W. B. Liu. Reproduction behavior in generative AI training: Copyright law evaluation from a market perspective. Journal of Gansu University of Political Science and Law, 1-19[2026-04-30].
[45] C. Y. Guan. Justification and rule improvement for including commercial use of data training in fair use. Journal of Hainan University, 1-11[2026-05-30].
[46] W. Du. Analysis of copyright extended collective management system. Intellectual Property, 2013, (1): 62-65+74.
[47] S. W. Mei, Y. Gong. Constructing copyright incentive mechanisms for hierarchical classification of data training use. China Publishing Journal, 2025, (20): 68-72.
[48] H. J. Lu. On market-oriented copyright collective management. Electronic Intellectual Property, 2007, (3): 16-20.
[49] S. J. Bao, D. M. Xiao. Copyright law responses to generative AI training data: EU exception rules and implications for China. Library Tribune, 2025, 45(6): 123-132.
[50] J. X. Zhang. Research on interest-sharing mechanisms for work data source providers in AI training. Intellectual Property, 2025, (5): 111-126.
[51] K. X. Ruan, X. Y. Huang. Research on copyright issues in generative AI data training. China Copyright, 2024, (5): 61-72.
[52] G. Z. Wang. Institutional analysis and legislative construction of copyright opt-out implied license. Contemporary Law Review, 2015, 29(3): 106-112.
[53] W. Du. Analysis of copyright extended collective management system. Intellectual Property, 2013, (1): 62-65+74.
[54] S. W. Mei, Y. Gong. Constructing copyright incentive mechanisms for hierarchical classification of data training use. China Publishing Journal, 2025, (20): 68-72.
[55] W. J. Zhang. On copyright regulation pathways for data used in big model training. East China Law Review, 2025, (2): 79-92.
[56] J. X. Zhang. Research on interest-sharing mechanisms for work data source providers in AI training. Intellectual Property, 2025, (5): 111-126.