美国法院于2024年12月20日批准了Anthropic与作家集体诉讼案的和解方案，Anthropic将向原告支付15亿美元和解金[1]。该案件自2024年8月至12月20日结束[1]，涉及Anthropic使用盗版书籍数据集训练商业模型Claude的行为。
根据诉讼内容，Anthropic在训练Claude模型时使用了包含近20万本Books3电子书、700万本LibGen和PiLiMi平台的盗版书籍[1]。此外，Anthropic还曾购买纸质书进行破坏性扫描以建立数据库[1]，尽管这一做法符合美国"首次销售原则"，但仍引发伦理批评[1]。
根据法院判决，91.3%的被侵权作品已被申领，每本书籍可获得3000美元的和解金[1]。原告的律师费初期要求为3亿美元，最终被法院削减至1亿美元[1]。该案件在社交媒体上获得了2000多万次浏览量[1]，被视为人工智能行业版权侵权的重要判例。
Anthropic has agreed to pay $1.5 billion in settlement to resolve a class-action lawsuit filed by writers and copyright holders over the use of pirated books in training its Claude AI model [1]. The case concluded on December 20, 2024, after proceedings that began in August of that year [1].
The dispute centered on Anthropic's use of unauthorized datasets containing approximately 200,000 books from Books3, as well as 7 million pirated titles from LibGen and PiLiMi platforms, in developing its commercial AI system [1]. Additionally, the company came under scrutiny for purchasing physical books and subjecting them to destructive scanning to build its own database, a practice that, while technically compliant with U.S. first-sale doctrine, drew ethical criticism [1]. The settlement requires Anthropic to distribute compensation to affected authors, with 91.3% of copyrighted works already claimed by claimants receiving $3,000 per title [1]. The court reduced the plaintiffs' attorney fees from an initially requested $300 million to $100 million [1].
The verdict is considered a landmark ruling in copyright infringement cases within the artificial intelligence industry [1], generating significant public attention with over 20 million views across social media platforms [1].
