A New Precedent: Anthropic's Historic Copyright Settlement
The AI world was shaken by the news that Anthropic, the company behind the advanced AI model Claude, is facing a colossal $1.5 billion settlement over copyright infringement claims. This case, which could become the largest copyright settlement in history, signals a monumental shift for the artificial intelligence industry, placing a bright spotlight on the often-murky world of AI training data.
The Core Allegations: Piracy for Progress
The class-action lawsuit, filed in the U.S. District Court for the Northern District of California, was brought forth by a collective of authors and copyright holders, including Andrea Bartz. The central claim is that Anthropic knowingly used pirated digital libraries, such as Library Genesis and Pirate Library Mirror, to train its Claude AI models.
The settlement, if approved, will set a precedent for AI companies paying for their use of pirated websites.
This isn't just about a single company's misstep; it's about establishing a legal framework that could hold the entire industry accountable. The plaintiffs argued that Anthropic engaged in large-scale copyright infringement by downloading and commercially exploiting hundreds of thousands of books without permission, all under the defense of "fair use."
The Court's Stance: "Inherently, Irredeemably Infringing"
The court's view on the matter was unequivocal. It rejected Anthropic's "fair use" argument, describing the company's use of pirated materials as "inherently, irredeemably infringing." The investigation revealed the immense scale of the operation, with plaintiffs' counsel reviewing hundreds of thousands of documents and inspecting at least three terabytes of training data to prove their case.
The Billion-Dollar Price Tag
Anthropic's recent financial success puts the staggering settlement amount into perspective. The company recently closed a $13 billion Series F funding round at a post-money valuation of $183 billion. For its investors, the $1.5 billion settlement might be viewed as simply the cost of doing business in a rapidly evolving legal landscape.
The payment structure is detailed and demanding:
- Minimum Payout: $1.5 billion, plus accrued interest estimated to be as high as $126.40 million.
- Per-Work Escalation: If the list of infringed works exceeds 500,000, Anthropic must pay an additional $3,000 for each new work added.
- Payment Schedule: The settlement will be paid in installments following preliminary and final court approval.
Beyond the Money: The Settlement's Strict Terms
The consequences for Anthropic extend far beyond the financial penalty. The settlement agreement includes several critical, non-monetary terms:
- Data Destruction: Anthropic is legally obligated to destroy all original files and copies of the works obtained from the pirated libraries within 30 days of the final judgment.
- No Future Immunity: The settlement only covers past infringements. It does not protect Anthropic from future lawsuits related to copyright infringement or claims arising from the outputs of its AI models. If Claude generates content that is substantially similar to a copyrighted work, the company can be sued again.
The Ripple Effect on the AI Industry
This case serves as a watershed moment, starkly contrasting with previous legal battles like the Google Books case, where Google scanned legally purchased physical copies. Anthropic's direct use of pirated digital content crossed a clear legal line.
The verdict sends shockwaves through the AI community, especially for companies like OpenAI, which is currently facing its own lawsuit from The New York Times. The era of leveraging gray-area data to build foundational models is coming to an end. AI companies will now need to invest heavily in legally licensed data, which will dramatically increase the cost of model development. This shift will not only reshape business strategies but will likely impact the pricing of AI services for consumers and businesses worldwide.
標題：Anthropic 的 15 億美元震撼彈：重新定義 AI 未來的版權訴訟
摘要：AI 巨頭 Anthropic 因使用盜版書籍訓練其 Claude 模型而面臨歷史性的 15 億美元和解。這一里程碑式的案件樹立了一個關鍵先例，迫使 AI 行業正視其數據來源的法律和財務後果，并從根本上改變了開發基礎模型的經濟模式。
正文：
新的先例：Anthropic 的歷史性版權和解
人工智能界因一則消息而震動：先進 AI 模型 Claude 的開發公司 Anthropic，正面臨一場高達 15 億美元的版權侵權和解。此案可能成為歷史上金額最高的版權和解，標誌著人工智能行業的一次巨大轉變，也讓 AI 訓練數據這個通常模糊不清的領域受到了前所未有的關注。
核心指控：為進步而盜版
這起集體訴訟由包括 Andrea Bartz 在內的多位作家和版權持有人在美國加州北區地方法院提起。其核心指控是，Anthropic 故意使用 Library Genesis 和 Pirate Library Mirror 等盜版數字圖書館來訓練其 Claude AI 模型。
該和解協議若獲批准，將為 AI 公司使用盜版網站付費樹立一個先例。
這不僅僅關乎一家公司的過失，更在於建立一個可能讓整個行業承擔責任的法律框架。原告方認為，Anthropic 在未經許可的情況下，下載並商業化利用了數十萬本書籍，構成了大規模的版權侵權，而其「合理使用」的辯護理由並不成立。
法院的立場：「內在的、不可救藥的侵權」
法院對此事的看法十分明確。法庭駁回了 Anthropic 的「合理使用」論點，將其使用盜版材料的行為描述為**「內在的、不可救藥的侵權行為」**。調查揭示了此次侵權的巨大規模，原告律師審閱了數十萬頁文件，並檢查了至少 3TB 的訓練數據來證實其指控。
十億美元的價碼
從 Anthropic 近期的財務成功來看，這筆驚人的和解金額就顯得不那麼難以承受。該公司最近以 1830 億美元的投後估值，完成了 130 億美元的 F 輪融資。對於其投資者而言，這 15 億美元的和解金或許只是在快速變化的法律環境中經營的必要成本。
付款結構詳細且嚴格：
- 最低賠償：15 億美元，外加預計高達 1.264 億美元的應計利息。
- 按作品數量增加：如果侵權作品清單超過 50 萬部，每增加一部，Anthropic 必須額外支付 3000 美元。
- 付款時間表：和解金將在法院初步和最終批准後分期支付。
金錢之外：嚴格的和解條款
對 Anthropic 而言，後果遠不止於金錢懲罰。和解協議還包括幾項關鍵的非金錢性條款：
- 銷毀數據：Anthropic 在法律上有義務在最終判決後的 30 天內，銷毀所有從盜版圖書館獲取的作品原始文件和副本。
- 未來無豁免權：該和解僅涵蓋過去的侵權行為，並不保護 Anthropic 免於未來的版權侵權訴訟，或因其 AI 模型輸出內容而引發的索賠。如果 Claude 生成的內容與受版權保護的作品高度相似，該公司仍可能再次被起訴。
對 AI 行業的連鎖反應
此案是一個分水嶺，與以往的法律戰（如谷歌圖書案）形成鮮明對比——谷歌當時掃描的是合法購買的實體書。Anthropic 直接使用盜版數字內容的行為，則明確越過了法律紅線。
這一判決給整個 AI 社區帶來了衝擊，尤其是像 OpenAI 這樣正被《紐約時報》起訴的公司。利用灰色地帶數據來構建基礎模型的時代即將結束。未來，AI 公司將需要投入巨資購買合法授權的數據，這將大幅增加模型開發的成本。這種轉變不僅將重塑商業戰略，還可能影響到面向消費者和企業的 AI 服務定價。