US AI Copyright Ruling: Implications for IP Law

Copyright Protection for Content Used in AI Training: A Recent Ruling

Recent legal developments have potentially strengthened copyright claims leveled against companies utilizing artificial intelligence.

The Thomson Reuters vs. Ross Intelligence Case

A U.S. federal judge issued a summary judgment last week in the dispute between Thomson Reuters, a major technology company, and Ross Intelligence, a legal technology firm.

The court determined that Ross Intelligence’s practice of employing Reuters’ content for the purpose of training its AI-powered legal research platform constituted an infringement upon Reuters’ intellectual property.

Potential Impact on Ongoing AI Copyright Litigation

This decision may carry significant weight in the context of the over 39 copyright lawsuits pertaining to AI that are presently being processed within the U.S. legal system.

However, it doesn't automatically guarantee success for plaintiffs asserting that AI firms have violated their IP rights.

Nuances of the Ruling

While a favorable development for copyright holders, the ruling isn’t a definitive victory for all parties bringing similar claims.

Each case will likely be evaluated on its own merits, considering the specific details of content usage and the nature of the AI application.

The legal landscape surrounding AI and copyright remains complex and is actively evolving.

Understanding the Headnote Controversy

Ross, an AI company, faced allegations of utilizing headnotes – concise summaries of judicial rulings – sourced from Westlaw, a legal research platform owned by Reuters. The AI was promoted as a system capable of analyzing legal documents and facilitating searches within court records.

Ross contended that its application of copyrighted headnotes was permissible under the law due to its transformative nature. This argument posited that the headnotes were being repurposed for a distinctly different function and targeted at a new market.

However, Judge Stephanos Bibas, overseeing the case, expressed skepticism regarding this claim in his summary judgment. He found Ross’s justification unpersuasive.

According to Judge Bibas’s assessment, Ross was essentially re-presenting Westlaw headnotes in a manner that mirrored Westlaw’s core legal research services. The platform, he concluded, did not introduce novel interpretations, insights, or commentary, thereby weakening Ross’s assertion of transformative use.

Commercial Intent and its Impact

Bibas’s ruling also highlighted Ross’s commercial objectives as a factor diminishing the validity of its defense. The startup aimed to generate revenue from a product that directly challenged Westlaw’s offerings.

This pursuit of profit occurred without substantial “recontextualization” of the intellectual property protected by Westlaw, further undermining Ross’s position.

Shubha Ghosh, a professor specializing in intellectual property law at Syracuse University, characterized the outcome as a “strong victory” for Thomson Reuters.

“While the trial is still pending, Thomson Reuters has secured a summary judgment, representing a significant win at this juncture of the legal proceedings,” Ghosh explained. “The judge also confirmed that Ross did not qualify for summary judgment based on defenses like fair use and merger, meaning the case will proceed to trial with Thomson Reuters holding a considerable advantage.”

• The case centers around the use of copyrighted legal summaries.
• Ross argued for transformative use, which the court rejected.
• Commercial motivations played a role in the judge’s decision.

Limited Scope of Application

Currently, at least one group of plaintiffs involved in a separate AI copyright case has requested a court to consider Judge Bibas’ ruling. However, it remains uncertain if this precedent will influence other judicial decisions.

Bibas’ assessment specifically differentiated between “generative AI” and the AI system utilized by Ross, which did not produce original content but rather presented pre-existing judicial opinions.

Generative AI, central to copyright litigation against entities like OpenAI and Midjourney, is commonly trained on extensive datasets sourced from publicly accessible online content. Through exposure to numerous examples, this type of AI can produce outputs including speech, text, images, videos, and musical compositions.

Many companies involved in the development of generative AI contend that fair use principles protect their practice of data scraping for training purposes, without providing compensation or attribution to the original data creators. They maintain that publicly available content is permissible for training and that their models generate transformative works.

However, this perspective is not universally accepted by copyright holders. Some highlight the issue of “regurgitation,” where generative AI replicates content closely mirroring its training data.

Randy McCarthy, a U.S. patent attorney from Hall Estill, indicated that Bibas’ emphasis on the “effects on the market for the original work” could be crucial for copyright holders pursuing legal action against generative AI developers. He also noted the relatively limited scope of Bibas’ opinion and the possibility of appeal.

“It is now established, at least within this case, that simply utilizing copyrighted material as training data for AI does not automatically constitute fair use,” McCarthy explained to TechCrunch. “However, this represents a single engagement in a broader conflict, and further developments are needed to establish legal precedents regarding the use of copyrighted materials in AI training.”

Mark Lezama, a litigation partner at Knobbe Martens specializing in patent disputes, believes Bibas’ ruling may have broader ramifications. He suggests the judge’s rationale could apply to various forms of generative AI.

“The court dismissed the fair-use defense because Ross leveraged [Thomson Reuters] headnotes to create a competing legal research platform,” he stated. “While the court acknowledged potential differences with generative AI scenarios, a news organization could reasonably argue that using its articles to train a generative AI model is comparable, as the AI then competes with the news source for audience engagement.”

Essentially, publishers and copyright owners engaged in disputes with AI companies have a modest basis for optimism following this decision – though it is a modest one.