Scaling Laws: AI Copyright Lawsuits with Pam Samuelson

The Lawfare Podcast & University of Texas School of Law | September 19, 2025
Host: Alan Rosenstein (Associate Professor of Law, University of Minnesota; Research Director, Lawfare)
Guest: Pam Samuelson (Richard M. Sherman Distinguished Professor of Law, UC Berkeley School of Law)

Episode Overview

This episode explores the rapidly evolving legal landscape at the intersection of artificial intelligence (AI) and copyright law. Host Alan Rosenstein speaks with IP law expert Pam Samuelson about the recent wave of copyright lawsuits involving AI companies such as Anthropic and Meta, recent pivotal court rulings, and the U.S. Copyright Office's controversial report on AI and fair use. The discussion covers core copyright doctrines, transformative use, market harm, remedies, and policy questions shaping the future of creative industries in the AI era.

Key Discussion Points & Insights

1. The Stakes of AI and Copyright Law (03:48–09:15)

• Legal Background: Rosenstein and Samuelson set the stage by explaining why copyright issues matter for AI, particularly with large language models (LLMs) and generative systems.

  • Copyright Owners’ Rights: Copyright grants creators exclusive rights, especially the right to reproduce their work; using works for AI training creates prima facie infringement, unless covered by fair use.

    "When people make fair uses, then even if it was a prima facie infringement, it's not an actual infringement..." — Pam Samuelson (04:34)

  • Fair Use Factors: The discussion introduces the four statutory fair use factors, with most debates centering on 'transformativeness' and effect on the market.

  • Balancing Incentives: The U.S. sees copyright as ultimately serving the public by incentivizing creation; the major fear is AI undermining creators’ incentives, leading to a "nightmare scenario" where new works dry up.

    "Hanging over all of these AI debates is this kind of nightmare scenario where AI essentially displaces all of this incentive to create new copyrighted work because you can’t make money anymore..." — Alan Rosenstein (08:10)

  • Empirical Uncertainty: Samuelson argues that while these fears are prominent, there isn't clear evidence yet that AI is destroying market demand for new works.

2. Supreme Court Framework: Warhol v. Goldsmith (11:35–19:49)

• Case Recap: The 2023 Supreme Court case involved Andy Warhol’s use of photographer Lynn Goldsmith’s work; ultimately, the Court found that licensing Warhol’s “Orange Prince” portrait for magazine use was not fair use because it substituted for the original’s commercial market.

  • The "transformative use" question is central but complicated; market substitution now weighs heavily.

  • Practical effect: Courts may be stricter regarding potential for lost economic opportunity due to unauthorized transformative uses.

    "[Courts] are going to really squint quite hard and look for any potential substitution effect, and that will weigh reasonably heavily on fair use analysis." — Alan Rosenstein (18:24)

  • Samuelson’s Take: She downplays "doom and gloom" about a tightening doctrine, noting that post-Warhol decisions have still allowed a range of traditional fair uses (e.g., documentaries using short clips).

3. Current Copyright Litigation: Anthropic & Meta Cases

Bartz v. Anthropic (22:48–33:01)

• Court’s Holding (Judge Alsup):

  • Major Ruling: Using copyrighted works as training data for AI models is fair use if it's highly transformative and non-expressive:

    "Using in-copyright works as training data for constructing a model for a generative AI system–that’s fair use because it’s transformative, highly transformative." — Pam Samuelson (22:48)

  • Authors’ Market Control: The judge found that authors do not have a right to control the market for transformative (training data) uses of their works.

  • Acquisition of Content: Digitizing lawfully acquired books is also fair use.

  • Contrast – Pirated Books: Judge Alsup ruled against Anthropic for using pirated book copies as training data.

    "He certified the class as all of the copyright legal or beneficial owners of copyright in books that had, number one, been registered with the Copyright Office and number two, had an ISBN or an Amazon number associated with it." — Pam Samuelson (34:28)

  • Settlement Issues: The proposed $1.5B settlement (about $3,000 per book) is controversial since it may benefit publishers more than the suing authors; the judge is skeptical about fairness.

  • Remedy Limitations: Due to practical concerns about crippling AI companies, courts likely won't impose injunctive relief that would destroy models; damages seem more likely.

Kadre (Cadre) v. Meta (42:27–47:30)

• Outcome (Judge Chhabria):

  • Found in favor of Meta; dismissed claims based on unauthorized training data (including pirated books).
  • Focused on Meta’s technical guardrails to prevent verbatim regurgitation of copyrighted content; no evidence of direct market harm.
  • Key Legal Theories:
    • Market for licenses of training data: Rejected as a right the authors control.
    • Market dilution theory: Raised as a possible avenue (if AI output meaningfully displaces demand for original books), but plaintiffs did not present sufficient evidence.
    • Chhabria’s Signal: Other plaintiffs might succeed by focusing their evidence on indirect but substantial market substitution (i.e., "market dilution”).

  "[Judge Chhabria] agrees...that the licensing market for uses of works as training data is just a market that the plaintiffs don’t have any right to control. Period." — Pam Samuelson (45:10)

• Conceptual Confusion: There’s legal uncertainty over how courts should treat alleged indirect market harm absent substantial similarity in expression—a point of skepticism for both host and guest.

4. Legal Remedies and Policy Solutions (39:52–55:46)

• Collective Licensing Models: Europe’s copyright system allows for collective licensing—potentially a model for U.S. reform, though transaction costs are a challenge for individualized licensing at scale.
• Remedies in Play: The real-world infliction of massive damages or operational restrictions on AI developers is unlikely, as judicial and market reluctance to "destroy the AI industry" is strong.
  • Statutory Damages: In code-related cases like Doe v. GitHub, statutory damages for copyright management info removal can be immense, shaping legal strategies.

5. Circuit Court & Policy Developments

Thomson Reuters v. Ross Intelligence (Third Circuit; 55:46)

• Key Issue: Using large numbers of legal headnotes to train AI tools—does this constitute fair use or market substitution? The district judge found infringement, influenced by the Warhol case, but was ambivalent and invited appellate review.

U.S. Copyright Office Controversy (56:14–57:53)

• Draft Report: The Copyright Office released a nuanced report on AI and fair use, finding that “some of these uses may be fair uses, and some...may not.” It introduced the “market dilution” concept.
• Political Turmoil: Unprecedented chaos erupted when President Trump tried to fire Register of Copyrights Shira Perlmutter; a court reinstated her, pending further appeals. This power struggle adds to the uncertainty around federal copyright policy.
  • The Biden administration’s AI policy platforms have conspicuously omitted copyright, deepening policy ambiguity.

6. Notable Quotes & Memorable Moments

• On fundamental copyright limits with AI:

  “What is different about this is the notion that something that is not substantially similar in expression...but that has a lot of the same information—well, copyright doesn’t protect information.” — Pam Samuelson (30:02)

• On why authors may never be able to effectively license training data rights:

  “There isn’t such a thing in the world, but let’s just say I did that. How much does every single author of every single book get? And how would you figure out what’s a fair compensation?” — Pam Samuelson (48:04)

• On prospects for big change soon:

  “It’s going to be a while...Nobody’s planning to make summary judgment motions in any of the other generative AI cases so far as I can tell, until 2026...” — Pam Samuelson (50:38)

Timeline of Important Discussion Segments

| Timestamp | Segment/Topic | |-------------|---------------------------------------------------------------------| | 03:01–04:34 | Intro to copyright law and fair use in the AI context | | 06:07–08:10 | Transformativeness, substitution, and incentive theory | | 11:35–19:49 | Warhol v. Goldsmith: Implications for transformative use | | 22:48–33:01 | Bartz v. Anthropic decision and transformative fair use | | 34:28–39:52 | Settlement, damages, and remedy constraints in AI copyright cases | | 42:27–47:30 | Kadre v. Meta and the rise of market dilution as a legal argument | | 50:38–55:46 | Upcoming appeals and policy outlook | | 56:14–57:53 | Copyright Office report & D.C. political controversy |

Conclusion

This episode tracks a decisive shift in how U.S. courts and policymakers are approaching AI and copyright. With early court decisions (notably Bartz v. Anthropic and Kadre v. Meta) trending in favor of AI model developers on the core issue of using copyrighted materials for training, unresolved tensions remain around market harm, possible new doctrines like market dilution, and appropriate remedies. Meanwhile, policy uncertainty and political drama (especially at the Copyright Office) keep the copyright landscape unsettled. Listeners are left with an expert’s caution: major developments and possible appellate or legislative change are likely years away.

Contact the show: scalinglaws@lawfaremedia.org
Find more content: lawfareblog.com