生成式AI對著作權之衝擊──以日本規範為中心

The impact of generative AI on copyright－focusing on Japanese regulations

由於生成式AI的快速發展，對著作權法制造成極大衝擊，日本因意識到此趨勢，陸續發布多個文件及指引，以建構AI創新與著作權保護平衡之法制，本文以日本提出之指引與文件為主要研究對象，探討AI訓練資料是否得主張合理使用、AI生成物之著作權歸屬、AI生成物侵權可能樣態與避免方法，以及潛在之侵權責任認定等問題。
本文首先介紹日本著作權法所保護之著作物定義，並介紹日本著作權法第30條之4對於「非以享受為目的」之運用得無須經著作權人同意及可使用著作之特殊規定，後針對AI開發訓練階段使用著作是否得主張合理使用進行討論，並討論AI生成與利用在何種情形下會侵害著作權，及AI生成物是否受到著作權保護等問題進行討論。最後討論我國實務目前對上述問題之見解，即可如何參酌日本之指引與文件，建構我國AI與著作權之法制。

Generative AI has developed at a pace that places unprecedented pressure on established copyright doctrine. Anticipating these stresses, Japan has released a succession of guidelines and policy papers designed to calibrate the balance between safeguarding copyrights and encouraging AI-driven innovation. This paper analyzes those Japanese materials as its principal reference point and engages four core questions: whether and on what grounds training data usage can be justified; how (and if) copyright subsists in AI-generated outputs and who, if anyone, owns it; what infringement patterns may emerge in the production and subsequent exploitation of AI outputs and how such risks can be mitigated; and how liability should be allocated where infringement is found. The analysis first clarifies the notion of a protected “work＂ under Japanese law and unpacks Article 30-4, which authorizes certain non-enjoyment uses without rights holder consent. It then scrutinizes the legal character of copying at the model development phase, probes the protectability of AI outputs, and maps potential infringement scenarios. The final section turns to Taiwanese practice, suggesting how Japan's evolving framework can inform the design of a local regime governing AI and copyright.