Court Rules Against AI Use of Copyrighted Content in Thomson Reuters Case

In a significant ruling, a US court has established that utilizing copyrighted material without consent to train artificial intelligence (AI) does not qualify as "fair use" under copyright laws. This decision is primarily favorable to copyright holders, as the judge ruled against the AI company involved in the case.

The case features Thomson Reuters, the owner of the well-known Reuters news service, and Ross Intelligence, a now-defunct company that provided access to a database of court cases via machine learning technology.

Back in 2020, Thomson Reuters initiated a lawsuit against Ross Intelligence, claiming that Ross had scraped content from its law database, Westlaw. Thomson Reuters argued that this scraping constituted copyright infringement.

Ross Intelligence pleaded that its use of Thomson Reuters' content should be shielded by the "fair use" provision in copyright law. However, on February 11, Judge Stephanos Bibas from the US District Court for the District of Delaware dismissed this argument in a recent ruling.

Judge Bibas had previously referred the copyright infringement issue and the "fair use" defense to a jury but later decided to withdraw that referral after a closer examination of the facts. In his summary judgment, he stated, "A smart man knows when he is right; a wise man knows when he is wrong," acknowledging his previous error.

Despite the ruling, the case is not completely resolved and will still be presented to a jury to address additional issues such as the validity of Thomson Reuters' copyrights and whether Ross copied specific systems used by Westlaw. However, the court has definitively rejected the fair-use defense.

This outcome initially appears to favor copyright holders because it sets a precedent against the use of copyrighted content for training AI, mirroring arguments made against AI developers by music industry stakeholders.

AI companies like Anthropic, Suno, and Udio have faced lawsuits from music companies over similar allegations. For instance, Anthropic argued in its defense against Universal Music Publishing and others that its Claude chatbot was trained on copyrighted song lyrics, which it allegedly reproduced upon request.

Suno and Udio have also incorporated the fair use argument into their defenses, admitting that their AI tools might have utilized copyrighted recordings for training. They contend that their methods are permissible.

While the Thomson Reuters decision theoretically favors copyright holders, it complicates matters when we consider the nuances of copyright law.

Understanding Fair Use in Copyright Law

The concept of fair use is meant to protect freedom of expression and foster innovation and learning. It often allows for parody, commentary, educational purposes, and some non-commercial research involving copyrighted works.

When courts evaluate whether unauthorized use of copyrighted content is permissible, they consider four key factors:

• Factor 1: The purpose and character of the use, specifically whether it is “transformative” in relation to the original work.
• Factor 2: The nature of the copyrighted work, where more creative works enjoy more protection than less creative ones.
• Factor 3: The amount and significance of the portion taken; using less of a work is generally less likely to be seen as infringement.
• Factor 4: The impact of the use on the potential market for the original work, essentially whether the new use diminishes its value.

In the Thomson Reuters case, Judge Bibas sided with Thomson Reuters on the first and fourth factors but ruled against it on the second and third. However, traditionally, the first and fourth factors carry more weight in court, which led him to favor Thomson Reuters overall.

Implications for Copyright Holders

Judge Bibas supported Thomson Reuters regarding the purpose and character of the use, stating that Ross's integration of Thomson Reuters’ data did not have a different purpose or character than the original work, as Ross aimed to compete with Westlaw. This finding aligns with the position of the music industry, where they can argue that AI-developed music lacks a transformative element, as it ultimately resembles the copyrighted music it utilized.

Under the fourth factor, the judge again ruled in favor of Thomson Reuters, concluding that Ross intended to develop a competing product and that they could have created their tool without infringing on Thomson Reuters' copyrights.

In contrast, factors regarding the nature of the work and the amount taken did not favor Thomson Reuters in the way they might in music copyright cases, as each song is subject to its unique copyright.

Challenges for Copyright Holders

The ruling brings good news for copyright holders, but with critical caveats. The judge explicitly noted that his decision does not extend to generative AI, which is at the center of ongoing disputes in the music industry.

Judge Bibas remarked that Ross’s AI was not generative and merely provided users with existing judicial opinions. This distinction is significant, as generative AI—like that employed by Anthropic, Suno, and Udio—constructs new content from learned data.

This difference in AI capabilities could lead to contrasting fair use assessments in future court cases involving music copyright claims. The courts may weigh the fairness of generative AI’s use of copyrighted works by determining whether the AI's outputs reduce the value of those original works or present a new transformative use.

As the music industry prepares legal strategies against generative AI, the crucial questions will revolve around whether these AI outputs are truly transformative and how they might affect the market dynamics regarding original music.

The Thomson Reuters case marks a crucial step toward defining copyright rights in the context of AI, yet many uncertainties and legal challenges remain before either side can claim an outright victory in this evolving landscape.

court, AI, copyright, music, fairuse