Navigating Fair Use in Generative AI: Insights from Recent Judicial Opinions

A Comparative Analysis of Bartz v. Anthropic and Kadrey v. Meta Platforms

Navigating the Frontiers of Fair Use in Generative AI: Recent Judicial Opinions

The legal landscape surrounding generative AI is evolving rapidly, with two recent judicial opinions poised to influence its future. The core issue at hand is whether using copyrighted works to train large language models (LLMs) constitutes fair use under US copyright law. While one ruling embraces a forward-thinking interpretation, the other stumbles, potentially undermining the progress made in AI development.

The Fair Use Framework

To understand the significance of these cases, it’s essential to grasp the framework established by the US Copyright Act for determining fair use. Courts analyze four key factors:

1. Transformative Nature of Use: Is the new use adding something novel or just repurposing the original work?
2. Nature of the Works: Are the original works more factual than creative? Have they been published for a long time?
3. Amount of the Original Used: How much of the original work was utilized in the new context?
4. Market Harm: Does the new use harm the original work’s market?

Both recent cases focus primarily on the first and last factors, underscoring their importance in evaluating AI’s burgeoning landscape.

The Right Approach: Bartz v. Anthropic

In the case of Bartz v. Anthropic, three authors challenged Anthropic for using their books to train its Claude chatbot. Judge William Alsup’s ruling supports what the Electronic Frontier Foundation (EFF) has articulated for years: using copyrighted works to train AI models is fair use. The judge highlighted the transformative nature of this process:

“Training gen-AI is ‘transformative—spectacularly so’ and any alleged harm to the market is purely speculative.”

Judge Alsup likened the training of AI to an aspiring writer reading various works—not to replicate but to innovate. He rejected the authors’ assertions that any model capable of generating competitive content constituted infringement. The court recognized that generative AI’s output transcends authors’ expectations of control over their work.

A Fumble on Fair Use: Kadrey v. Meta Platforms

In contrast, Kadrey v. Meta Platforms takes a less favorable stance on fair use. Here, authors sought to prevent Meta from using their works to train the Llama chatbot. Unfortunately, Judge Vince Chhabria’s ruling veers into overly speculative territory. His opinion focuses on what could have justified a ruling against Meta rather than factual evidence.

While the court ultimately ruled in Meta’s favor due to a lack of substantial evidence from the plaintiffs, it made critical errors along the way. The ruling broadly suggests that without licenses for all copyrighted works used in training, any AI operation could be deemed illegal. This assertion overlooks the transformative nature of AI training processes, suggesting instead that fair use would not commonly apply.

The ruling relies on three flawed assumptions:

1. Market Harm as a Primary Factor: It incorrectly positions potential market harm as the most pivotal point of fair use analysis, despite supreme court precedents indicating no single factor predominates.
2. Intentional Market Competition: The ruling assumes that AI developers design models to replicate and compete directly with the works used in training—an assertion the evidence did not support.
3. Misunderstanding Market Dilution: Legally, copyright does not protect against market dilution unless the new works are otherwise infringing. In creativity, competition is often a driving force, not a hindrance.

Looking Ahead

The current discourse around generative AI mirrors previous technology panics, where concerns often overshadow the innovations at play. Thoughtful interpretations of fair use, like that in the Bartz ruling, are essential to fostering environments conducive to creativity and technological advancement. Conversely, rulings like Kadrey suggest a cautious approach that could stifle innovation.

As courts grapple with the implications of generative AI, it is imperative they heed the framework established in Bartz. Balancing copyright protections with the need for transformative uses will not only safeguard creativity but also promote an era of unprecedented innovation. We stand at a crucial juncture, and how we navigate these legal challenges will shape the future landscape of artificial intelligence and copyright for years to come.