Unsealed Authors Guild filings allege OpenAI and Microsoft knew their book training data came from pirated LibGen copies
2026年9月17日 · 自动从公开来源采集，未经人工审核。
这篇内容目前只有 英语 原文版本。
What happened: On September 17, 2026, class plaintiffs in Authors Guild et al. v. OpenAI (No. 1:2023cv08292, S.D.N.Y., part of the consolidated MDL against OpenAI and Microsoft) filed a 162-page statement of undisputed facts and a memorandum in support of partial summary judgment. The unsealed papers allege OpenAI downloaded about 117,500 books from Library Genesis in 2018 and later torrented roughly 35TB matching LibGen's full collection; that Sam Altman disclosed LibGen use to Bill Gates and Microsoft CTO Kevin Scott in April 2019; and that OpenAI deleted its LibGen files in summer 2022 through an internal effort discussed in a Slack channel first named #excise-libgen ("Project Clear"). The filings quote internal messages, including then-Research Director Dario Amodei calling the LibGen set "a bit sketchier" and staff worrying about the "optics" of the practice surfacing publicly.
Why it matters to creators: These are the most detailed internal records yet made public about how a major AI company's book training corpus was assembled, including what executives knew and when. They center the dispute on the source of the training data rather than just its use.
What is not yet known: All of this is the plaintiffs' presentation of the evidence, not a judge's findings of fact. OpenAI and Microsoft argue fair use; no ruling on the summary judgment motion has issued, with a hearing expected in early 2027.