
Anthropic以15亿美元和解版权诉讼,AI训练数据合法性争议持续发酵
美国法院批准了针对Anthropic的史上最大版权和解案,同时田纳西大学另以专利侵权起诉该公司,凸显全球AI产业在数据使用规则上的深层法律冲突。
美国加州北区联邦法院法官本周批准了Anthropic与逾50万名作者达成的15亿美元集体诉讼和解协议,这是美国版权法历史上金额最高的和解案。同一天,田纳西大学研究基金会在特拉华州联邦法院起诉Anthropic,指控其AI系统侵犯了该校在神经网络和机器学习领域的两项专利。这两起案件将Anthropic推至AI产业知识产权争议的中心,也反映出大语言模型训练数据获取方式所面临的日益严峻的法律审视。
根据美国法院文件,作者团体指控Anthropic在2022年通过一个故意违反版权的数字档案,下载了超过700万册盗版书籍用于训练其聊天机器人Claude。负责该案的法官William Alsup在去年6月的裁定中认为,将书籍用于AI训练本身可能构成“合理使用”,但下载和存储盗版书籍的行为独立构成侵权。Anthropic副总法律顾问Aparna Sridhar表示,公司对91%的合格权利人已申领赔偿感到满意,并希望“了结此事”。然而,部分作者反对和解,认为每本书约3000美元的赔偿过低,且律师费过高,法官虽将律师费从申请的1.875亿美元削减至1.01亿美元,但仍驳回了这些反对意见。另有部分作者和出版商选择退出和解,单独提起诉讼,相关案件仍在推进。
在欧洲和俄罗斯,该案引发了更广泛的立法讨论。德国《南德意志报》援引法庭文件披露,Anthropic创始人曾在内部通信中将获取盗版书库称为“正好赶上”,此后公司又雇佣前Google图书项目负责人Tom Turvey,通过购买实体书、切除书脊、扫描后销毁的方式获取训练文本,并曾尝试向出版商获取授权但未成功。俄罗斯《公报》报道称,俄媒体公司、IT开发商和版权持有人团体已联合请求政府和执政党修改AI发展法案,要求为训练神经网络使用作品建立许可机制,反对现行草案中允许自由使用公开可访问内容的条款。这些动向表明,不同法域的政策制定者正试图在促进技术创新与保护创作者权益之间寻找平衡点。
对中国和亚洲而言,此案提供了重要参照。中国已施行的《生成式人工智能服务管理暂行办法》要求训练数据“具有合法来源”,但具体到版权作品的“合理使用”边界,司法实践仍处于探索阶段。美国法院将“下载存储”与“训练使用”区分为两个独立法律行为的逻辑,可能影响亚洲国家在界定AI数据合规时的思路。目前,Anthropic还需应对退出和解的作者诉讼以及田纳西大学的专利侵权指控,后者若胜诉,可能对依赖类似神经网络架构的AI模型产生更深远的技术限制。全球范围内,针对AI公司的版权和专利诉讼预计将持续增加,国际社会围绕AI训练数据规则的立法博弈也将进一步加速。
| 欧洲大陆媒体 | −0.80 | critical |
|---|---|---|
| 俄罗斯及独联体媒体 | +0.70 | aligned |
| 拉丁美洲媒体 | 0.00 | neutral |
| 东南亚媒体 | +0.30 | aligned |
Anthropic deliberately violated copyrights and even celebrated it. The settlement is an admission, but the practice must be stopped.
By quoting internal messages, the narrative creates the impression of intentional wrongdoing, strengthening moral condemnation.
The account omits that a court later ruled that training AI on books can be considered fair use, which relativizes the legal basis of the settlement.
Anthropic reached a settlement without admitting guilt, and the court confirmed that training AI on books is fair use. This is a victory for the AI industry.
Emphasizing the court's fair use ruling makes the settlement appear not as a punishment but as a compromise favorable to Anthropic.
The mention that Anthropic deliberately used a pirate archive and that the founder celebrated it is absent.
The $1.5 billion settlement is a milestone, but it does not resolve the tension between AI innovation and copyright. Fair use is still being tested in the courts.
By presenting multiple angles (settlement, patents, opinion), the coverage suggests the issue is complex and without a definitive conclusion, avoiding taking sides.
Neither the internal Anthropic email about the deliberate use of a pirate archive nor the court ruling that considered AI training as fair use are mentioned.
The Anthropic settlement is a minor bump on the road to an AI-powered future. The real story is the unprecedented progress AI will bring in the next five years.
By framing the legal case as a mere detail within a larger narrative of inevitable progress, the article minimizes its significance and directs attention to optimistic forecasts.
The article omits all specifics of the Anthropic copyright case, including the $1.5 billion settlement, the allegations of deliberate infringement, and the fair use ruling, thereby avoiding any negative framing.