UPDATE – AI and Copyright – Federal Appeals Court Affirms That Training an AI Research Tool on Copied Copyrighted Text is Copyright Infringement and Not a Fair Use

UPDATE – AI and Copyright – Federal Appeals Court Affirms That Training an AI Research Tool on Copied Copyrighted Text is Copyright Infringement and Not a Fair Use  

In an earlier post,we discussed a ruling from the District of Delaware regarding whether an AI-based legal research system that trained on copyrighted headnotes from Westlaw constituted copyright infringement and/or whether such training on copyrighted material was a fair use.  Because of the novelty of the issue, Judge Bibas immediately certified the questions for appeal to the United States Court of Appeals for the Third Circuit. 

One point to note is that judicial opinions are publicly accessible and would not typically be covered by any type of copyright, such that an AI tool could train on the judicial opinions.  In this instance, however, Ross's AI-based legal research system used the Westlaw "headnotes" that are not part of the judicial opinions, but added by Westlaw to assist in coding and organizing judicial opinions by subject or issue, to put it simply.  Thus, in the case, the issue was not whether copying or using the judicial opinions was permissible, but rather, whether training on the headnote content was copyright infringement, and if so, whether it was a fair use that provided a defense to copyright infringement. 

The Third Circuit affirmed Judge Bibas' opinion,and found that the case was "no more than an ordinary copyright case".  The Third Circuit determined that Westlaw's headnotes contained copyrightable subject matter.  The Third Circuit also determined that the process by which Ross's AI-based legal research system was trained and worked included a step where Ross (through third parties) had its system trained on memos addressing legal issues that users might query about, and (simplifying) the memos "used thousands of Westlaw headnotes" to construct their collectively content.  This was found to be copyright infringement. 

In terms of fair use, Title 17, United States Code, Section 107 states that a fair use determination shall be made based on at least the following factors: (1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work. 

Here, the Court determined that Ross was admittedly in commercial, for-profit competition with Westlaw.  The Court, however, found that the second fair use factor favored a finding of fair use (i.e., that the works were quite factual in nature to accurately convey the law).  For the third factor, the Court determined that even though Ross may have only copied a small portion of Westlaw's headnotes, the copying was not necessary to train Ross's AI system.  The Court also determined that what was copied was "the entire text of the 25,000 Westlaw-written headnotes" and the purpose was to create a commercial substitute for Westlaw, not some other, transformative purpose.  Thus, the third fair use factor weighed against a fair use.  As to the fourth factor, the Third Circuit determined that, "[t]he undisputed evidence here reveals that ROSS copied the headnotes to create a competing legal-research platform in the same market as Westlaw. '[U]nrestricted and widespread' copying of Thomson Reuters's headnotes to develop substitute legal research tools, like ROSS did, would harm Westlaw's position in that original market. … ROSS offers no evidence to rebut the alleged harm in this market."   

The Third Circuit concluded that 2,243 of the Westlaw headnotes were "original enough for copyright protection" and that Ross's "use of the headnotes was not fair." 

The key takeaway is that, if you are training an AI system on potentially copyrighted materials, you should consult with an experienced copyright attorney to determine whether the training is permissible or not.  The issue in the Thomson Reuters v. Ross case was that Ross created memos from Westlaw's headnotes, which made the infringement question more typical.  If material is not being copied prior to training, that is a different situation that needs to be separately assessed.  In this context, keep in mind that this is a quickly evolving area of law that is factually-intensive.  So, again, consult with an experienced copyright attorney before you put your AI system into motion.