Last week, the Delaware District Court issued a case of first impression ruling regarding the use of copyrighted works for artificial intelligence (AI) training. In Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., No. 1:20-CV-613-SB (D. Del. Feb. 11, 2025), the District Court reversed from a previous 2023 summary judgment opinion after the court “studied the case materials more closely and realized that [their] prior summary-judgement ruling had not gone far enough.”
Case Facts
In this case, Ross Intelligence had created a legal research search engine with the aid of artificial intelligence. To support usage of AI in legal research, Ross needed to establish a database with legal questions and answers. Thomson Reuters, being one of the biggest legal platforms in the world, was the primary choice for Ross to assemble their database. Ross thus attempted to license Thomson’s content for training purposes. Thomson refused their proposal, in large part because they are developing their own AI tool called CoCounsel.
Undaunted, Ross sought assistance elsewhere and landed upon LegalEase for training data. LegalEase provided Ross so-called “Bulk Memos,” approximately 25,000 of them. These memos contained the questions and answers that Ross needed. However, these memos also contained Westlaw headnotes, with specific instructions from LegalEase that lawyers using the memos should “not just copy and paste headnotes directly into the questions.” Using those memos, Ross built their database, including the Westlaw headnotes. Ultimately, Thomson Reuters sued Ross for copyright infringement.
Ruling
Based on these facts, the Court found that the Westlaw headnotes were copyrightable because they had a minimal level of creativity. District Court Judge Bibas analogized that the development of the headnotes was akin to creating a marble sculpture, where the artist creating the sculpture chooses what to cut down to create their image, identical to a legal expert reading a judicial opinion and “identifying which words matter and chiseling away from the surrounding mass expess[ing] . . . the important point of law from the opinion.” Further, the headnotes are considered to be a compilation due to Thomson’s selection and arrangement of its headnotes. With this established and after carefully analyzing through 2,830 headnotes, the Court ruled for Thomson finding that approximately 2,243 headnotes were copied and that there was substantial similarity between the 2,243 headnotes and the bulk memos.
In arguably the most important section of this decision, the Court denied all of Ross’s defenses to copyright infringement. which included innocent infringement, copyright misuse, merger, scenes à faire, and even fair use. The Court readily found that innocent infringement, misuse, and scenes à faire were inapplicable in this case. As to merger the Court also rejected the argument, noting that “there are many ways to express points of law from judicial opinions.”
The only worthwhile defense the Court treated in some detail was fair use. In doing so, the Court deliberated on its four elements: (1) the use’s purpose and character; (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 of potential market. The Court gave Ross the benefit of the doubt on factors 2 and 3. However, it found for Thomson Reuters on factors 1 and 4. Relying on precedent establishing that the first and fourth factors weigh most heavily in its analysis, the Court rejected Ross’s fair-use defense, granting summary judgment for Thomson Reuters.
Other Developments Around the World
This decision has set the tone for the plethora of ongoing cases that are faced with an identical core issue: using copyrighted work to train AI. It also joins the recent decisions seen around the world.
In Europe, for instance, the Municipal Court in Prague decided in S. Š. v. Taubel Legal, advokátní kancelář s.r.o., (Oct. 2023) that an image created through AI without significant human input cannot be copyright-protected. In this case, S. Š. used generative AI to create an image using the following prompt: “create a visual representation of two individuals signing a business contract in a formal setting, like a commercial room or a law office in Prague, showing only the hands.” The resulting image was published on the law firm’s website without S. Š.’s consent. The Court found that the work was not a unique result of the author’s creative intellectual activity under the Czech copyright law. However, the decision left open the possibility that the plaintiff, who generated the prompt to produce the AI image, could have possibly received copyright protection of the image had they included a more specific set of details or personalization in their prompt.
Further east in China, in Li Yunkai v. Liu Yuanchun, the Beijing Internet Court answered just that on November 2023: finding that an AI-generated image is copyrightable, and the author of that prompt is entitled to the authorship of that image. In this case, the plaintiff had developed over 100 prompts with various specifications to achieve the final image, generated by an open-source image generation software. The defendant had used the plaintiff’s image, removed their watermark, and posted the image alongside an article on a social media platform. The plaintiff sued for infringement. The court determined that the parameters set out in the 100 prompts dictated the characters, design, and presentation of the resulting image, reflecting the plaintiff’s personal and arrangement preferences. Thus, the Court found that the originality element under Chinese copyright law was satisfied.
Conclusion
The cases noted here definitely open the door for claims of copyrights in AI-generated material, particularly given the emphasis on the prompts used to generate the result. Future case decisions will likely come down to the quantity and quality of the prompts used to generate the results. Borrowing from Judge Bibas’ analogy in Thomsom Reuters, the prompts are akin to the chiseling the artist may do in creating the AI-generated work-product.
The information you obtain at this site, or this blog is not, nor is it intended to be, legal advice. You should consult an attorney for advice regarding your individual situation. We invite you to contact us through the website, email, phone, or through LinkedIn. Contacting us does not create an attorney-client relationship. Please do not send any confidential information to us until such time as an attorney-client relationship has been established.
You ought to be a part of a contest for one of
the most useful websites on the net. I will recommend this web site!
Thank you George. Much appreciate the compliment. It is hard to know if anyone reads these musings anymore.