加州北区联邦法院已批准Anthropic与版权作者达成的15亿美元和解协议。[1]该和解针对这家AI公司使用盗版书籍训练Claude模型的侵权纠纷，法官Araceli Martínez-Olguín于2026年7月20日签署了最终批准令。[1]和解覆盖约48.2万至50万部作品，相当于每部作品约3000美元的补偿，成为美国版权史上规模最大的AI相关和解。[1]
然而，法院的裁定揭示了这起纠纷的复杂性。[1]法官Alsup认定，AI模型的训练行为本身构成合理使用，真正的侵权在于Anthropic从盗版影子图书馆下载700多万本书籍的行为，被法院描述为"固有、不可救药的侵权"。[1]根据版权法故意侵权的最高法定赔偿标准——每部作品15万美元——理论敞口曾高达接近1万亿美元。[1]此外，法院已澄明该和解不构成约束性先例，这意味着类似案件的法律框架仍待确立。[1]
面对AI训练数据的版权问题，美国国会正在推进立法解决。[1]CLEAR法案（S.3813，2026年2月10日提出）要求AI开发者披露训练数据中每一件受版权保护作品的信息；TRAIN法案（H.R.7209/S.2455，2026年1月提出）则赋予版权人通过行政传票要求AI开发者披露作品使用情况的权利。[1]
The Northern District of California Federal Court approved a $1.5 billion settlement between Anthropic and copyright holders over the use of pirated books to train its Claude AI model [1]. Judge Araceli Martínez-Olguín signed the final approval order on July 20, 2026 [1]. The settlement covers approximately 482,000 to 500,000 works, translating to roughly $3,000 per title, marking the largest AI-related copyright settlement in American legal history [1].
Despite the settlement's magnitude, the underlying legal principles remain contested. Judge Alsup ruled that the training process itself constitutes fair use, but determined that downloading over 7 million books from piracy platforms constitutes "inherent and irreparable infringement" [1]. The theoretical maximum liability could have reached nearly $1 trillion, based on the statutory maximum of $150,000 per work for willful copyright violations [1]. The settlement also accounts for approximately $122 million in attorney fees and litigation costs [1], while 350 class members chose to opt out, involving 1,802 works [1], and the court received 54 objections [1].
The resolution does not establish binding precedent, leaving broader questions about AI training data governance unresolved [1]. Congress is advancing transparency legislation to address copyright issues in AI training: the CLEAR Act (S.3813, introduced February 10, 2026) would require AI developers to disclose every copyrighted work in their training data [1], while the TRAIN Act (H.R.7209/S.2455, introduced in January 2026) would allow copyright holders to issue administrative subpoenas demanding disclosure of how their works were used [1].