OpenAI盜版書籍評測：內部信曝高層知法犯法 | OpenAI Book Piracy Review: Emails Show Execs Knew

9月 28, 2026

By Kit 小克 | AI Tool Observer | 2026-09-28

🇹🇼 OpenAI盜版書籍評測：內部信曝高層知法犯法

OpenAI盜版書籍官司在2026年9月有了新進展：紐約南區法院解封了Authors Guild v. OpenAI／The New York Times v. Microsoft and OpenAI案的大量法庭文件，內容顯示OpenAI高層很早就知道用盜版書訓練GPT模型可能違法，卻選擇繼續執行。這起訴訟原告包括暢銷作家David Baldacci、John Grisham、George R.R. Martin、Jodi Picoult、Jonathan Franzen，指控OpenAI大規模用未授權書籍訓練模型，衝擊作家生計。

解封文件揭露了什麼？

根據解封的訴訟文件，2019年8月OpenAI內部有一則訊息寫著：「We trained GPT-3 on pirated stuff! No sharing that!」（我們用盜版資料訓練了GPT-3！別說出去！）直接點出公司內部早就清楚訓練資料的來源問題。

2020年政策主管示警：時任OpenAI政策主管Jack Clark在內部備忘錄中寫道，公司的工作將「逐漸取代人力勞動」，並預期「會有一群藝術家跳出來抗議我們在做的事，而我們很可能會無視他們的擔憂，照樣發布」。
「Project Clear」滅證疑雲：文件顯示OpenAI在2022年夏天啟動代號「Project Clear」的內部專案，研究副總裁Bob McGrew在信中寫道：「現在是把LibGen從我們系統和儲存空間裡清掉的好時機」，原告律師將此解讀為銷毀證據的行為。

為什麼這件事值得AI用戶關注

這起AI版權訴訟不只是作家和科技巨頭的恩怨，而是牽動每一個在用ChatGPT、Copilot的人：

如果法院認定訓練資料侵權成立，OpenAI可能被要求刪除受影響模型或支付鉅額賠償，間接反映到訂閱與API定價上。
案件會成為之後所有AI公司訓練資料授權的判例，Anthropic、Google、Meta都密切關注（各家也各自面對類似訴訟）。
對開發者來說，這是提醒：用AI生成內容做商業用途前，最好先確認你所在地區對「AI訓練資料合法性」的最新立場，避免下游踩雷。

案件現況與後續

目前案件仍在訴訟階段，尚未有最終判決，OpenAI與Microsoft也對指控提出反駁，主張訓練行為屬於合理使用（fair use）。但解封文件讓外界第一次看到內部溝通的原始樣貌，也讓「高層明知故犯」的敘事更難被公關話術蓋過去。

好不好用，試了才知道。

🇺🇸 OpenAI Book Piracy Review: Emails Show Execs Knew

The OpenAI book piracy lawsuit took a major turn in September 2026: a federal court unsealed a trove of documents in Authors Guild v. OpenAI and The New York Times v. Microsoft and OpenAI, revealing that OpenAI executives knew early on that training GPT models on pirated books could be illegal — and pushed forward anyway. The plaintiffs include bestselling authors David Baldacci, John Grisham, George R.R. Martin, Jodi Picoult, and Jonathan Franzen, who allege OpenAI trained its models on unauthorized copies of their books at scale.

What the Unsealed Documents Show

According to the unsealed filings, an internal OpenAI message from August 2019 read: “We trained GPT-3 on pirated stuff! No sharing that!” — direct evidence that the company was aware of the provenance problem with its training data from early on.

A 2020 warning from OpenAI’s own policy chief: Jack Clark, then OpenAI’s policy director, wrote in an internal memo that the company’s work would “increasingly lead to us creating systems that substitute for the labor of people,” and predicted “there will be a point where a bunch of artists express worry about what we’re doing here and we’ll likely ignore their concerns and release anyway.”
“Project Clear” and the deletion question: Filings describe an internal 2022 initiative called Project Clear, under which VP of Research Bob McGrew wrote that “now is the right time to excise Libgen from our systems and storage.” Plaintiffs’ lawyers argue this amounts to evidence destruction.

Why This AI Copyright Lawsuit Matters to Regular Users

This isn’t just a fight between authors and a tech giant — it touches anyone using ChatGPT, Copilot, or any downstream product built on these models:

If courts find the training data infringing, OpenAI could face model takedowns or major damages — costs that tend to flow back into subscription and API pricing.
The case will likely set precedent for how every AI lab licenses training data going forward; Anthropic, Google, and Meta are watching closely, and facing similar suits of their own.
For developers shipping AI-generated content commercially, it’s a reminder to check the current legal standing on training-data provenance in your jurisdiction before you build a business on top of it.

Where the Case Stands Now

The case is still in litigation — no final ruling yet — and OpenAI and Microsoft continue to argue their training practices qualify as fair use. But the unsealed documents give the public its first unfiltered look at internal communications, and make the “executives knew” narrative much harder to spin away.

好不好用，試了才知道。

Sources / 資料來源

延伸閱讀 / Related Articles

AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

搜尋此網誌

AI小貼士