精选85Rohan Paul行业动态多源精选 ×3
索尼音乐与华纳查普尔起诉Anthropic，索赔数十亿美元
Sony music and Warner chappell just sued Anthropic, with a multi-billion lawsuit…
推荐理由
AI从业者必看，这是版权诉讼的重大进展，直接牵涉Anthropic训练数据来源与合规风险，建议关注后续对模型训练实践的影响。
Sony music and Warner chappell just sued Anthropic, with a multi-billion lawsuit, calling it ‘one of history’s biggest and most blatant ongoing intellectual property thefts.’
索尼音乐和华纳查普尔刚刚起诉了Anthropic，索赔数十亿美元，称其是‘历史上最大、最明目张胆的持续知识产权盗窃之一。’
Sony Music Publishing and Warner Chappell want up to $150,000 per infringed work, disclosure of Anthropic's training data and collection methods, and destruction of infringing copies.
索尼音乐出版公司和华纳查普尔要求每件被侵权作品最高赔偿15万美元，公开Anthropic的训练数据及收集方法，并销毁侵权副本。
Some of the things, the complaint alleges
诉状中指控的一些内容如下：
- They say that Anthropic co-founder Benjamin Mann discussed acquiring LibGen data with CEO Dario Amodei, who approved the torrenting; an earlier court found Mann knew the books were pirated.
- That the pirated LibGen and PiLiMi material still reached commercial Claude models indirectly, through synthetic data and reinforcement feedback from other Anthropic models trained on that material.
- They also accuse Anthropic of stripping copyright-identifying information from training material and rewarding accurate reproduction of copyrighted lyrics during finetuning.
- Claude finetuning included copyrighted song lyrics, with reviewers selecting accurate lyric reproductions over inaccurate ones, effectively rewarding verbatim recall.
- 他们称，Anthropic联合创始人本杰明·曼与CEO达里奥·阿莫德伊讨论了获取LibGen数据的事宜，后者批准了种子下载；此前一家法院认定曼知道这些书籍是盗版的。
- 盗版的LibGen和PiLiMi资料仍通过合成数据和来自基于这些资料训练的其他Anthropic模型的强化反馈，间接进入了商业化的Claude模型。
- 他们还指责Anthropic从训练材料中剥离版权识别信息，并在微调过程中奖励对受版权保护歌词的准确复制。
- Claude的微调包含了受版权保护的歌曲歌词，评审人员选择准确的歌词复现而非不准确的，实际上是在奖励逐字记忆。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力