AI Copyright Law: Courts Define the Role of Fair Use ⁣in Model Training

Updated June 27, 2025

The legal⁢ landscape ⁢surrounding generative AI is ⁢rapidly evolving, with recent court⁢ decisions tackling the critical question of whether using copyrighted material to ‍train AI models constitutes fair use. These rulings are poised to significantly shape the⁢ future of AI development, clarifying the role of copyright in this emerging field.

At the heart of these‍ cases lies ⁢the question: Is it lawful under fair use principles ⁢to utilize copyrighted works when training Large Language Models (LLMs) for AI chatbots? The U.S. Copyright Act provides a framework⁢ for answering⁤ this, requiring courts to consider several factors, including whether the use is transformative, the nature of the copyrighted works, the amount used, and ⁢the potential harm to the market for the original work. Judges in recent cases have primarily focused on the transformative nature of the use and its potential impact on the market.

In ⁣ Bartz v.Anthropic, authors sued Anthropic, alleging their books ⁤were used to train the Claude chatbot. Judge William Alsup sided with Anthropic, affirming that fair use protects the use of copyrighted works for AI training. he emphasized that this training is “transformative” and⁤ that ⁢any alleged market harm remains ⁣speculative. The court likened the process to creating search engines, where copying books to create a new, transformative LLM is protected.

“[U]sing⁣ copyrighted works to train LLMs to generate new text was quintessentially transformative…[LLMs] trained upon works not to race ahead and⁤ replicate or supplant them-but to turn a hard‍ corner and create something different.”

The Bartz ruling also‍ dismissed claims that AI models capable of ⁢generating content similar in theme or style to existing works are inherently infringing.⁤ The court recognized ⁣that building AI models that create new works falls outside the scope of copyright control.

Though, the subsequent ruling in Kadrey v. Meta Platforms ‍ took a different approach.In this case, ⁢authors sued Meta, the developer of the ‘Llama’ chatbot, arguing ⁣that fair use did not apply. ⁢While the court ultimately ruled in Meta’s favor, Judge Vince Chhabria’s opinion raised‍ concerns about potential “market⁢ dilution,” suggesting that training AI without licenses‍ could be “illegal” in many cases.

The court asserted that fair use ‍might not apply to AI training due to hypothetical scenarios ‍where AI-generated works ‍compete with and diminish the value of the original works used for training. This theory rests on the premise that AI developers would train models⁤ on specific types of ⁢works to generate content in the same genre, directly competing with the originals. The Kadrey ruling acknowledges a lack of ⁢evidence supporting this scenario.

Furthermore,the court’s opinion suggested that market harm is the most significant factor in determining fair⁢ use,a notion contradicted by ⁢the ⁢Supreme Court’s stance in Campbell v.Acuff-Rose. Copyright law aims to foster new expression