返回档案库

案例库 · 法务与合规 · 法务决策 · 2020-2025

这条还没译成中文,下面是英文原文。

Ross Intelligence trained its AI on Westlaw's headnotes — a court called it theft

Thomson Reuters refused to license Westlaw's editorial headnotes, so the legal-AI startup got them through a vendor anyway — and lost the company defending it.

Ross Intelligence · Thomson Reuters · 2025-02-11

怎么回事

Ross Intelligence built a competing AI legal-research tool and wanted to train it on Westlaw's headnotes — Thomson Reuters' editorial summaries of case law. Thomson Reuters refused to license them, so Ross paid a third-party vendor, LegalEase, to produce 'Bulk Memos' built directly from Westlaw's headnotes and Key Number System.

Thomson Reuters sued for copyright infringement in 2020. On February 11, 2025, Judge Stephanos Bibas of the District of Delaware ruled against Ross on summary judgment, finding the headnotes copyrightable and Ross's use not fair use — the copying wasn't transformative, and it created a competing product in Thomson Reuters' own potential AI-training-data market.

The ruling was the first major US decision on fair use in AI training, and it went the copyright holder's way on every factor the court weighed. Ross had already shut down years earlier, unable to keep funding the defense — the case survived its own defendant.

为什么会这样

  • Thomson Reuters refused to license Westlaw's headnotes for AI training, so Ross obtained them through a vendor instead.
  • The court found 2,243 headnotes copied and rejected fair use on the market-harm factor — training an AI on them competed directly with Thomson Reuters' own AI-licensing market.
  • It became the first major US ruling on AI-training fair use, and it favored the copyright owner on every factor.
  • Ross Intelligence shut down before the ruling arrived — the legal defense outlasted the company's ability to pay for it.
代价company shut down defending the suit灾难级

教训

Being refused a license is not a green light to get the same material through a vendor — a court will read that as the theft it functionally was.

后来呢

The ruling became the reference point for later AI-copyright suits against larger AI labs, cited for its finding that intermediate copying for training is not automatically transformative.

资料来源

发现哪里写错了?告诉我们。

Comments · 0

    登录 后就能评论。

    类似的案例

    这家公司栽倒的地方,别处有人漂亮地解开过。 第二意见 →