加利福尼亚州圣何塞 —— Anthropic 公司于 10 月 21 日敦促加利福尼亚州的一名法官裁定，该 AI 公司利用数百万首歌词训练其聊天机器人 Claude 的行为，受美国版权法中的“合理使用”条款保护，以此寻求驳回环球音乐集团及其他音乐出版商提起的诉讼。
SAN JOSE, California – Anthropic urged a judge in California on Oct 21 to rule that the AI company’s use of millions of song lyrics to train its chatbot Claude is protected as “fair use” under US copyright law, as it seeks to scuttle a lawsuit brought by Universal Music Group and other music publishers.
美国加州北区联邦地区法官李友美（Eumi Lee）在圣何塞举行的听证会上未当庭作出裁决，但就合理使用问题向双方发问，该问题可能成为 AI 时代数十起其他高风险版权侵权案件的关键。
US District Judge Eumi Lee did not rule at a hearing in San Jose but pressed both sides on the fair use question, which could be key to dozens of other high-stakes copyright infringement cases in the AI era.
李法官还质疑出版商方的主张，即 Claude 创作歌词的能力威胁到出版商歌词的市场。
Lee also questioned the publishers’ assertion that Claude’s ability to compose lyrics threatens the market for the publishers’ lyrics.
Anthropic 方律师、威尔默·黑尔律师事务所的 Sonal Mehta 辩称，其 AI 训练促成了变革性新内容的创作，完全符合合理使用原则。
Anthropic lawyer Sonal Mehta of WilmerHale argued that its AI training leads to the creation of transformative new content, putting it squarely under the fair use doctrine.
出版商方律师、Oppenheim + Zebrak 律师事务所的 Matt Oppenheim 回应称，Anthropic 滥用其歌词生成竞争性内容，损害了人类词曲作者的利益。
The publishers’ attorney Matt Oppenheim of Oppenheim + Zebrak responded that Anthropic misuses its lyrics to generate competing content that harms human songwriters.
Oppenheim 称，在六个月期间内，Claude 收到了近 12 万次生成“合成”歌词的请求。他表示，像 Claude 这样的大语言模型“很容易用 AI 生成的内容淹没市场”。
Oppenheim said Claude received nearly 120,000 requests to generate “synthetic” lyrics over a six-month period. He said large language models like Claude “could overwhelm the market with AI-generated content very easily”.
Mehta 反驳称，AI 并未降低对歌词的需求，且出版商方无法指出 Claude 或其他大语言模型造成的任何具体市场损害实例。
Mehta countered that AI has not reduced demand for the lyrics, and that the publishers could not identify any specific instances of market harm from Claude or other large language models.
Mehta 还表示，UMG 自己曾称 AI “绝大程度上是音乐行业的净利好”。
Mehta also said that UMG itself has said that AI is “overwhelmingly a net positive” for the music industry.
早期合理使用裁决 合理使用可能决定作家、新闻机构、视觉艺术家等起诉 Anthropic、OpenAI 和 Meta Platforms 等科技巨头就其 AI 训练提起的待决诉讼的命运。
Early fair use rulings Fair use could determine the fate of pending lawsuits brought by authors, news outlets, visual artists and others against tech giants including Anthropic, OpenAI and Meta Platforms over their AI training.
另外两位北加州法官已就这一备受关注的问题作出裁决，但意见不一，即生成式 AI 训练总体上是否构成对受版权保护材料的合理使用，这将使科技公司免于因在未经内容所有者许可的情况下使用数百万受版权保护的作品来训练其系统而承担的责任。
Two other Northern California judges who have already weighed in on the closely watched issue have split on whether generative AI training generally makes fair use of copyrighted material, which would shield tech companies from liability for using millions of copyrighted works to train their systems without content owners’ permission.
现已退休的美国地区法官威廉·奥尔苏普于 2025 年支持 Anthropic 的观点，认为其 AI 训练对书籍的使用属于合理使用。两天后，美国地区法官文斯·查布里亚在一起类似诉讼中裁决 Meta 胜诉，但警告称 AI 训练“在许多情况下”不构成合理使用。
Now-retired US District Judge William Alsup sided with Anthropic in 2025 on its argument that its AI training made fair use of books. Two days later, US District Judge Vince Chhabria ruled for Meta in a similar lawsuit but warned that AI training “in many circumstances” would not be fair use.
在李法官审理的案件中，音乐出版商环球音乐、康科德和 ABKCO 于 2023 年起诉 Anthropic，指控其使用他们的歌词训练 Claude 以响应人类提示，侵犯了他们包括碧昂丝、滚石乐队和海滩男孩在内的艺术家数百首歌曲的版权。
In the case being heard by Lee, music publishers UMG, Concord and ABKCO sued Anthropic in 2023, alleging its use of their lyrics to train Claude to respond to human prompts infringed their copyrights in hundreds of songs by artists including Beyonce, the Rolling Stones and the Beach Boys.
Anthropic 辩称，其 AI 训练对歌词进行了变革性使用，旨在“帮助 Claude 理解人类语言，并推动科学、商业和教育领域的进步与生产力”。
Anthropic has argued that its AI training makes transformative use of song lyrics to “help Claude understand human language and enable progress and productivity in science, business and education”.