Matthew Sag is one of the leading scholars on copyright and generative AI — the questions of whether training AI models on copyrighted works is infringement or fair use, when model outputs infringe, and how the law should allocate the risks. He came to these questions with an unusual head start: he had been writing about the copyright status of machine copying since 2009, more than a decade before ChatGPT.
The core contributions
Copyright Safety for Generative AI, 61 Houston Law Review 295 (2023), proposed one of the first frameworks for how AI developers can reduce the risk of infringing outputs, and introduced the “Snoopy problem”: the tendency of generative models to memorize and reproduce distinctive protected characters — the more abstractly a character is protected, the more likely a model is to reproduce it. Fairness and Fair Use in Generative AI, 92 Fordham Law Review 1887 (2024), assesses the fair use claims of AI developers, distinguishing training uses that are genuinely non-expressive from uses that compete with the works they were trained on.
The Globalization of Copyright Exceptions for AI Training, 74 Emory Law Journal 1163 (2025) (with Peter K. Yu), maps the international landscape: how the United States, the European Union, Japan, Singapore, and other jurisdictions are converging on — and diverging over — copyright exceptions for AI training. Copyright’s Jagged Frontier, 76 Duke Law Journal (forthcoming 2026), continues the project as the litigation wave matures.
The courts are engaging this work directly: in Kadrey v. Meta Platforms, 788 F. Supp. 3d 1026 (N.D. Cal. 2025), the first major fair use ruling on LLM training data, Judge Chhabria adopted the “indirect substitution” framing from Fairness and Fair Use in Generative AI in analyzing market dilution — while rejecting the article’s conclusion on the fourth factor. Being the position a court must engage to decide the question is its own measure of influence.
In the policy arena
- U.S. Senate testimony — In July 2023, Sag testified before the Senate Judiciary Subcommittee on Intellectual Property at its hearing on artificial intelligence and copyright.
- U.S. Copyright Office — Comments and reply comments in the Office’s Notice of Inquiry on Artificial Intelligence and Copyright (2023, with Pamela Samuelson and Christopher Jon Sprigman); invited presentations in the Office’s international AI webinar and listening sessions.
- White House OSTP — Copyright and the AI Action Plan (2025), a submission on copyright policy for the U.S. AI Action Plan.
- Courts — Brief of Copyright Law Professors as Amici Curiae in Thomson Reuters v. Ross Intelligence (3d Cir. 2025), the first appellate battleground over AI training and fair use.
- Internationally — Invited addresses and briefings for policymakers and industry in Australia, Korea, Argentina, and the European Union.
Commentary as it happens
Sag writes regularly about the AI copyright litigation and policy landscape on this site — recent analyses include the Disney–OpenAI deal, cherry-picked memorization evidence, the Books3 dataset, and copyright pressure on Wikipedia — and in venues such as ProMarket (The False Hope of Content Licensing at Internet Scale, 2025). His casebook, Copyright Law in the Age of AI, integrates these questions into the teaching of U.S. copyright law.
Related
Non-expressive use · Text and data mining · Fair use · Impact & influence