Source: Mondaq

Patent specifications and drawings published by the China National Intellectual Property Administration become publicly accessible technical resources, but they may also meet the originality requirement of the Copyright Law and thus qualify as protected works. Formal drafting and administrative examination do not automatically convert such documents into administrative materials; courts have found originality in wording, sentence arrangement and drawing design and have protected them as literary or graphic works in cases such as (2021) Jing 73 Min Zhong No. 4384 (Beijing Intellectual Property Court) and (2022) Shan Zhi Min Zhong No. 112 (Shaanxi High People’s Court). This judicial recognition coexists with the disclosure purpose of patents, which limits the exercise of copyright where reproduction and dissemination serve legitimate technical-information needs without harming the normal exploitation of the work.

The lawfulness of including published patent materials in AI training raises potential reproduction and information‑network communication rights issues under Copyright Law and must be assessed against the three‑step test (statutory circumstance, non‑conflict with normal exploitation, and no unreasonable prejudice to rights holders). The Shanghai Intellectual Property Court’s 2023 LoRA/Altman decision introduced “analytical use” as a possible fair‑use rationale for generative AI training while stressing provider responsibilities to prevent infringing outputs. Policy instruments such as the Interim Measures for the Administration of Generative Artificial Intelligence Services require lawful data sources and transparency, and industry practices vary: some AI patent‑tool developers rely on large public datasets but adopt record‑keeping, internal review, and contractual allocations of responsibility to mitigate copyright risk. Future clarity is expected from further judicial precedents and regulatory frameworks to balance IP protection and technological innovation.