The Cost of Training Generative AI: How Bartz v. Anthropic’s Potential Class Action Settlement Will Impact the AI Industry and the Legal Field

Sept. 26, 12:13 PM

The world is accelerating towards widespread adoption and integration of artificial intelligence (AI) into all aspects of daily life. While one sector of society views AI as the most transformative technology of the 21st century, many artists and designers view generative AI as an existential threat to their livelihood. A proposed class action settlement from the Northern District of California looks to ease those fears.

Bartz v. Anthropic Background

In August 2024, Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson (“Plaintiffs”) filed a class action lawsuit against Anthropic in the Northern District of California. The Plaintiffs alleged Anthropic engaged in largescale copyright theft by training its large language model (LLM), Claude, with pirated versions of their copyrighted works.

On June 23, 2025, Judge Alsup granted summary judgment for Anthropic, stating its use of copyrighted works to train LLMs and generate new text was “quintessentially transformative” and thus constituted fair use. In the same order, Judge Alsup denied summary judgment for Anthropic’s use of pirated books in its central library and ordered a trial to determine Anthropic’s potential liability for piracy.

By mid-July, Judge Alsup granted an order in favor of the motion for class certification. The order defined the class as all beneficial or legal copyright owners whose books were uploaded to the online pirate libraries Library Genesis (LibGen) or Pirate Library Mirror (PiLiMi) and downloaded by Anthropic. After the class certification, Anthropic faced potential statutory damages exceeding hundreds of billions of dollars.

A Landmark Settlement Proposal

On August 26, 2025, the parties filed a notice of a proposed class-wide settlement. Ten days later, the parties filed a motion for preliminary approval of settlement. The principal terms of the settlement were:

  • Anthropic will pay the class at least $1.5 billion dollars plus interest. Based on an estimated 500,000 works included in the class, this amounts to roughly $3,000 per work
  • Anthropic will destroy the LibGen and PiLiMi datasets after the expiration of any litigation preservation or other court orders
  • Anthropic will be released for conduct up to August 25, 2025. Claims arising out of conduct after August 25 will not be released by the settlement

Simply put, this settlement is “the largest publicly reported copyright recovery in history, larger than any other copyright class action settlement or any individual copyright case litigated to final judgment.”

“How do we respect and reward human creators without impeding technological progress?”

Settlement Postponed

Despite the fanfare surrounding the proposed settlement, Judge Alsup denied the preliminary approval, citing concerns with the claims process and undue influence by two trade associations – the Authors Guild and the Association of American Publishers. The judge postponed the preliminary approval hearing until September 25 and requested the parties provide written answers to a series of questions (1, 2).

Legal and Policy Implications

The settlement poses potential legal and policy implications such as:

(1) Updating Copyright Law – As posed by the former Register of Copyrights, Shira Perlmutter, “How do we respect and reward human creators without impeding technological progress?” The Copyright Office has released a three-part report looking to answer that question. Now, Congress must do the hard work of updating copyright law to ensure it balances innovation and the rights of copyright holders.

(2) Settlement Framework – As highlighted by Professor Edward Green, Susman Godfrey’s use of its Shadow Library legal strategy helped it secure the largest settlement in U.S. copyright history without relying on fair use arguments. Parties in other pending copyright suits could view this framework as a viable option to quickly settle cases without going to trial. Additionally, this settlement framework may push the AI industry to establish legitimate data pipelines, accelerate the development of a mature licensing market for AI-training data, and promote collective licensing organizations.

(3) Saving an Industry but Creating Monopolies – Approving this settlement will signal a viable legal pathway to resolving similar claims. However, only Big Tech companies and well capitalized startups could pay billions of dollars in damages. This means scrappy startups face another barrier as they look to break into the generative AI space. Thus, the settlement could have the unintended consequence of stifling innovation and further consolidating Big Tech’s influence and power over the industry.

This case, though momentous, does not bind other districts to follow its analysis. However, it could be a foundational order for future decisions altering the development of the AI industry and copyright law.

Jalen Saunders

Jalen is a 2L at the University of North Carolina School of Law. Before law school, Jalen attended the Georgia Institute of Technology where he graduated with a B.S. in Nuclear & Radiological Engineering and a M.S. in Materials Science Engineering. He enjoys listening to music, working out, and spending time with his wife.