Context: Copyright enforcement against AI providers is now a global phenomenon (January 8, 2025 ai fray article). In the United States, where roughly three dozen AI copyright infringement lawsuits are pending, the “fair use” defense is particularly key (January 24, 2024 ai fray article).
What’s new: The United States District Court for the District of Delaware has just handed down a decision that constitutes the most important (though ultimately appealable) victory of a copyright holder over an AI provider. ThomsonReuters, the owner of the Westlaw legal research database service, has prevailed on its direct infringement claims against AI provider Ross, whose service competes with Westlaw, and has overcome the latter’s “fair use” defense. The decision was made by Third Circuit Judge (and Trump appointee) Stephanos Bibas (Wikipedia page) who is sitting on the district court by designation for the purpose of presiding over this landmark case. He originally denied summary judgment over those key questions, but based on further evidence postponed the trial in order to reopen the summary judgment proceedings that have now indeed served to narrow the case for trial purposes.
Direct impact: While some secondary questions regarding the validity of the asserted copyrights remain for the jury, this partial summary judgment is the next best thing to full summary judgment entering liability. No matter which side prevails in district court (most likely, ThomsonReuters), this case is very likely to be appealed to the Third Circuit where Judge Bibas’s peers will review his decision as he cannot become involved again at the next level of proceeding.
Wider ramifications: While many copyright plaintiffs suing AI providers will cite to this decision and argue that their cases warrant the same outcome, it is important to understand that fair use requires a highly fact-intensive inquiry and no two cases are identical. Also, Judge Bibas decided two of the four fair use factors in the defendant’s favor, which shows that it was not a slam dunk for the right holder. Any other defendant to an AI copyright action will have to argue, more than anything, that its service does not compete with the right holder’s own business the way Ross sought to displace Westlaw. For the avoidance of doubt, fair use is a U.S.-specific defense. Even to the extent that other jurisdictions have scope for similar theories, they tend to be more favorable to right holders anyway.
Here’s the partial summary judgment order:
This is going to become one of the most widely-read decisions in a U.S. copyright case.
Apart from the outcome, it is also interesting from a case management perspective that the judge presiding over the matter was hesitant at first to find that there was no material dispute of fact, only to change mind later and grant leave for renewed summary judgment motions.
Judge Bibas stopped short of entering a liability finding, but “the only remaining factual issue on liability is that some of those copyrights may have expired or been untimely created.” It is hard to imagine that none of the asserted copyrights will get past this hurdle, so the trial will primarily be about damages.
The questions that have been resolved prior to trial are direct infringement (yes) and various defenses raised by Ross, which have been thrown out: innocent infringement,
copyright misuse, merger, scenes à faire, and fair use. The last one is the single most important one, by a wide margin in fact.
Like in this case, it makes sense for other U.S. judges presiding over AI copyright cases to resolve fair use themselves. The Supreme Court showed the way in Oracle v. Google. Fair use is a mixed question of fact and law, thus does not have to be put before a jury.
The order summarizes the four statutory fair use factors as follows:
(1) the use’s purpose and character, including whether it is commercial or nonprofit; (2) the copyrighted work’s nature; (3) how much of the work was used and how substantial a part it was relative to the copyrighted work’s whole; and (4) how Ross’s use affected the copyrighted work’s value or potential market.
ThomsonReuters won (1) and (4), Ross won (2) and (3), but the final multifactorial balancing resulted in a clear win for the right holder. In the end, what proved outcome-determinative was
- the low creativity threshold for copyright protection (the decision notes that “[i]f taking 300 words out of President [Gerald] Ford’s memoirs could count as taking the heart of the work, so too can taking several thousand headnotes out of Westlaw);
- Ross’s failure to persuade Judge Bibas of its work having been transformative (in other words, “do it with AI” is not sufficient to establish the transformative nature of a derivative work); and
- the fact that what Ross offers competes with Westlaw as a direct substitute as opposed to serving a fundamentally different purpose.
The decision is not binding beyond the District of Delaware, and even there its precedential significance is limited as each case is different. Most AI copyright cases are pending in the Southern District of New York (Second Circuit) and the Northern District of California (Ninth Circuit), where a Delaware decision is out-of-circuit authority.
It is a huge win for ThomsonReuters, but does not weaken, much less gut, the fair use defense with a view to other pending U.S. copyright cases over the way copyrighted works are used to train AI systems.