Thursday, April 16, 2026
Executive Summary
Judge M. Margaret McKeown of the Ninth Circuit delivered the BTLJ Spring Lunch Keynote on May 7, 2026, arguing that generative AI presents copyright’s most profound doctrinal stress test since the internet, but that the 1976 Copyright Act’s flexible language—particularly the fair use factors and the undefined scope of “original works of authorship”—remains capable of absorbing the disruption, even as courts await a first appellate decision and Congress remains largely dormant.
Instructor(s)
Hon. M. Margaret McKeown, Senior Circuit Judge, United States Court of Appeals for the Ninth Circuit
Molly Van Houweling, UC Berkeley Law
Erik Stallman, UC Berkeley Law
Keywords
1976 Copyright Act generative AI fair use • transformative use doctrine AI training data • Bartz v. Anthropic fair use settlement • Kadrey v. Meta LLM training copyright • human authorship requirement AI-generated works • vicarious and contributory copyright infringement internet • DMCA safe harbor platform liability • Naruto v. Slater copyright authorship non-human • Does AI training on copyrighted books constitute fair use? • What does the 1976 Copyright Act say about artificial intelligence and authorship? • compulsory licensing regime generative AI copyright • market harm fourth fair use factor AI
Legal Analysis
The 1976 Copyright Act’s Dance with the Internet: Scale, Intermediary Liability, and the Limits of Judicial Adaptation
The story of copyright and the internet, Judge McKeown emphasized, is fundamentally “one of scale.” The network did not invent copying or distribution; it redefined their magnitude, velocity, and geographic reach in ways that forced courts to adapt doctrine never designed for a decentralized, borderless, and essentially free-to-use publishing infrastructure. The first collision came in 1993 in Playboy Enterprises, Inc. v. Frena, 839 F. Supp. 1552 (M.D. Fla. 1993), a bulletin-board case that McKeown described as deceptively simple but doctrinally consequential: the court found infringement without requiring proof of intent or knowledge, a reading that implied strict liability for direct infringement. Almost simultaneously, Religious Technology Center v. Netcom On-Line Communication Services, Inc., 907 F. Supp. 1361 (N.D. Cal. 1995), reached the opposite result, holding that a passive platform operator that merely enabled uploads could not be directly liable. These two decisions, McKeown observed, “set the stage for what is this continuing dance between the courts and Congress”—a choreography that produced the Digital Millennium Copyright Act of 1998, which Congress drafted with explicit reference to both cases. When peer-to-peer networks collapsed the distinction between distributor and infrastructure provider, the Ninth Circuit held in A&M Records, Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001), that facilitating mass infringement could ground both contributory and vicarious liability, and that no fair use defense was available. The Supreme Court extended that logic in Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005), introducing an inducement theory that pushed software liability doctrine further still. Most recently, Sony Music Entertainment v. Cox Communications, Inc. (4th Cir. 2023) overturned a $1 billion jury verdict and, in McKeown’s assessment, “froze the vicarious liability doctrine” by rejecting the proposition that an ISP’s mere knowledge of user infringement, without more, establishes willfulness—though the decision left numerous subsidiary questions unresolved. Taken together, McKeown argued, the internet cases suggest that the act’s principal structural challenge lay not in defining what counts as copying but in allocating liability across a layered ecosystem of intermediaries operating at unprecedented scale.
Generative AI and the Fair Use Fault Lines: Transformative Use, Market Harm, and the Rejection of AI Exceptionalism
The approximately eighty generative AI copyright cases now pending in federal courts have, as yet, produced no appellate decision—a reality McKeown described plainly as “the world that we operate in.” Of the district court decisions that do exist, two illuminate the emerging doctrinal battleground with particular clarity. In Bartz v. Anthropic PBC, No. 3:23-cv-03223 (N.D. Cal.), Judge William Alsup declined to rule that all AI training is categorically transformative, opting instead to sort the defendant’s conduct into “buckets” based on the provenance of the ingested works: lawfully purchased and digitized books qualified for a fair use analysis, while books downloaded from pirate sites did not. McKeown found the decision notable for what it rejected as much as for what it embraced. Alsup characterized generative AI as “quintessentially transformative” and compared LLM training to a reader aspiring to become a writer—reasoning by analogy, as McKeown observed courts habitually do when confronting novel technology—but he flatly rejected Anthropic’s argument that the existence of a pirated source library was irrelevant to the fair use calculus. “There is no kind of free-floating entitlement to copy first and justify later,” Alsup wrote, a formulation McKeown cited as evidence that the decision “rejected AI exceptionalism in the world of copyright.” The $1.5 billion settlement mooted any Ninth Circuit review. In contrast, Kadrey v. Meta Platforms, Inc., No. 3:23-cv-03417 (N.D. Cal.), remains pending, with Judge Vince Chhabria having found at summary judgment that Meta’s use of books from shadow libraries—including Library Genesis and Anna’s Archive—constituted fair use on the record before him, while explicitly cautioning that this holding did not establish the general legality of LLM training. Chhabria emphasized that LLM training combines two previously unseen features simultaneously: deeply transformative use of the work and the potential to flood existing markets with competing outputs. The doctrinal significance, McKeown observed, is that the first and fourth fair use factors—as construed by the Supreme Court in Harper & Row, Publishers, Inc. v. Nation Enterprises, 471 U.S. 539 (1985) (designating market effect “the most important” factor); Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021) (reaffirming fair use as “an equitable rule of reason”); and Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023) (narrowing the transformativeness inquiry)—will determine whether AI training survives judicial scrutiny. McKeown cautioned that the question of what kind of market harm copyright law is prepared to recognize remains, for now, entirely open.
Authorship, Institutional Actors, and the Future of Copyright Governance in the Age of AI
Beyond the infringement and fair use questions lies a deeper constitutional and statutory uncertainty about authorship itself. The 1976 Act deliberately left original works of authorship undefined, entrusting the contours of that phrase to judicial elaboration—a choice McKeown identified as more consequential than the drafters may have anticipated. The Copyright Office’s response to Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018)—which held that a macaque monkey lacked statutory standing to assert copyright—was to treat the decision as establishing the outer boundary of the human authorship requirement. McKeown observed that this requirement may now be “more load bearing than we originally thought,” because the emergence of AI-generated content forces courts to ask precisely what kind of human creative agency copyright law is designed to honor. The institutional landscape through which answers will emerge has also grown considerably more complex than the drafters of the 1976 Act anticipated. McKeown predicted that in the AI context “the courts will be first movers and not Congress,” given that forty-four copyright bills introduced since 2022 have all failed to advance. Congress’s most recent major legislative intervention was the Orrin G. Hatch–Bob Goodlatte Music Modernization Act of 2018. Executive Order 14179, issued in 2025 by the Trump administration, pushed toward a minimally burdensome national AI framework and characterized state AI laws as obstructive to innovation; yet thirty-eight states have already enacted some form of AI-specific legislation, raising preemption questions McKeown predicted will soon reach the courts. Europe is an additional structural variable: the European Union’s 2019 Directive on Copyright in the Digital Single Market established text-and-data-mining exceptions, and the EU AI Act subsequently layered transparency, safety, and opt-out obligations onto general-purpose model providers, giving European copyright law what McKeown described as an “ex anti regulatory and compliance dimension” that will shape global developer practices. On the question of amicus briefs—likely to proliferate in these cases—McKeown was direct: the most useful submissions explain the policy ramifications of a ruling across the statute as a whole, illuminate implications of the technology beyond the immediate record, and present arguments that fall within the ambit of the parties’ positions but that the parties themselves did not fully develop. A “me too” brief, she cautioned, adds nothing. McKeown concluded with a structural observation that echoed throughout the keynote: the act’s terms—originality, authorship, work, copy, and market—are being asked “to do new work in different factual circumstances that were previously truly unimagined by the drafters,” and it is the courts’ necessary task to interpret those definitions flexibly enough to accommodate innovation without converting every technological shift into a legislative emergency.
Generated by AI based on the Interview/Transcript below.
Key Takeaways
- Courts as first movers on AI. McKeown predicted that courts, not Congress, will set the initial doctrinal framework for generative AI and copyright, given that forty-four copyright bills since 2022 have all failed—”Congress will be a more leisurely follower.”
- No appellate AI copyright decision yet. As of the keynote, not a single appellate court has ruled on the merits of a generative AI copyright claim, leaving district court opinions as the only available precedent.
- Rejection of AI exceptionalism. Judge Alsup’s ruling in Bartz v. Anthropic rejected the proposition that AI developers hold a “free-floating entitlement to copy first and justify later,” applying standard copyright doctrine rather than creating a new AI-specific immunity.
- Source provenance matters for fair use. Alsup bifurcated the training-data analysis by legal source: lawfully purchased and digitized books could qualify as fair use; books downloaded from pirate sites could not—a distinction with enormous practical consequences for AI developers.
- First and fourth fair use factors are decisive. Both Bartz and Kadrey confirm that the transformativeness inquiry (factor one) and market harm (factor four, identified as most important in Harper & Row) will be the primary battlegrounds in AI copyright litigation.
- Human authorship is load-bearing doctrine. The undefined scope of “original works of authorship” in the 1976 Act and the Copyright Office’s reliance on Naruto suggest that the human authorship requirement will become a central and contested issue as AI-generated content proliferates.
- State laws will shape federal litigation. Thirty-eight states have enacted AI-specific legislation; while these laws may not directly address copyright, McKeown cautioned they will generate evidentiary records, disclosure obligations, and preemption disputes that will influence copyright cases.
- Effective amicus briefs serve the statute, not the party. McKeown advised that the most useful amicus submissions explain how a ruling would ripple across the Copyright Act as a whole, provide superior explanations of the underlying technology, and surface arguments within the parties’ positions that the parties themselves did not raise—not merely echo a party’s brief.
- Legislative history still functions, indirectly. Despite the turn toward textualism, McKeown noted that judges frequently cite cases that themselves cite legislative history, meaning, as she observed of Abbe Gluck’s scholarship, “legislative history lives on, but in secret.”
- The Act’s language remains generative. McKeown concluded that the 1976 Act’s enduring capacity to generate debate fifty years on—its terms being asked to “do new work” in unimagined factual circumstances—is itself evidence of the statute’s success, not its obsolescence.
B-CLE Recording (CLE: FREE) | Youtube Recording | Resource(s) | Speaker Bio(s) & Contact Info
Download the interview/transcript and slides here!
Interview/Transcript
This interview/transcript was based on a conversation on April 16, 2026 about 29th Annual BTLJ-BCLT Spring Symposium: Origins, Evolution, and Possible Futures of the 1976 Copyright Act, hosted by Berkeley Center for Law & Technology, UC Berkeley School of Law. The topic on “How Generative AI Is Reshaping Fair Use and Authorship Under the 1976 Copyright Act” was presented by keynote speaker Hon. M. Margaret McKeown, Senior Circuit Judge, United States Court of Appeals for the Ninth Circuit; introduced by Molly Van Houweling, UC Berkeley Law; and moderated by Erik Stallman, UC Berkeley Law.
Molly Van Houweling 00:15
I’m back and I am again, delighted to kick off one of the highlights of this conference, I’ll be introducing today’s lunchtime keynote speaker, The Honorable Judge Margaret McKeown, and preview I’m a big fan, a fan of her work across her legal practice, her judicial practice, her work at the ALI. She has served since 1998 as a judge on the US Court of Appeals for the Ninth Circuit, where she has authored many of our favorite copyright opinions and some that people might beg to differ with. We’ll give you an opportunity to talk about cases like Garcia versus Google, Hokin versus Skyy Spirits, Mange versus Maya, Dr Seuss Enterprises versus Comic Mix and Daniels versus Walt Disney, about the color coded emotion characters in the movie Inside Out. She has served really too many leadership roles to mention, but I will mention a few. She’s a member of the American Academy of Arts and Sciences, the counsel of the American Law Institute and an advisor to the restatement of copyright, where I’ve had the pleasure of interacting with her for over a decade now. And she’s past president of the Federal Judges Association. Before her appointment to the bench, she was the first woman partner at Perkins Cooley in Seattle and Washington, DC. She is widely renowned as a pioneering and globe trotting lawyer, jurist mentor to law clerks and to lawyers and to judges around the world. And I can think of no one better to speak on the topic. She has chosen today a view from the bench technology and the 1976 Copyright Act. Judge McKeown plans to speak for about 30 minutes, and then my Berkeley Law colleague Erik Stallman will moderate a question answer session with you the audience. So gather your thoughts in preparation for that, and please join me in a very enthusiastic welcome for Judge McKeown.
Hon. M. Margaret McKeown 02:19
Thank you. Well, thank you. Of course, it’s an honor to be here, and I think the organizers have done a marvelous job. I have really loved the panels before. I’ve actually been involved in copyright for about 50 years. So that’s just a fraction of copyrights history, but it coincides with the lifetime of the 76 Act. And of course, we’re celebrating the Golden Jubilee. Might also be interested to know that this is the Golden Jubilee for Apple, and also Six Flags, Great America. So we share quite a lot, but what really strikes me in looking back at the act is that the drafters really did understand with somewhat unusual clarity the challenges of writing copyright for a future that really couldn’t be clearly seen. So Congress was aware, as we know from the ACT language, that it was going to apply to these new technologies that were now known or later developed. So in that the act represented what I call this dance between Congress and the courts. So one of the students questions actually presaged that theme. And then we’ve gone on to say that there’s, there’s a lot of discussion today about both criticism and commentary about the courts, the Second Circuit, the Ninth Circuit, my decisions and other decisions, and I have to say that I do welcome it, and the kind of commentary and criticism that is here is very respectful, and I have to say it’s a far cry from some of the things I receive over the internet based on some of my other opinions. So I’m very happy to be here, but next time, I’m going to bring a whole phalanx of judges. Not to defend ourselves, but to have and perhaps that’s a different conference, and that’s the dance, perhaps, between the Academy and the courts. And of course, by the time of the 76 Act, the courts had really pushed copyright to its limits in an interpretive sense. And that really represents, I think a fairly standard division of labor as Justice Holmes once remarked, a judicial inquiry investigates, declares and enforces liabilities present or past. That’s the purpose and that’s the end. But legislation, on the other hand, looks to the future and changes and making a new rule. So amid really hundreds of landmark decisions in the copyright area, Congress has amended the act 79 times. Some amendments were major, like the DMCA, others, smaller but significant. But just think about this in this really exponentially explosive technological world we live in now, the last major legislation was eight years ago, 2018 and that was the Hatch-Goodlatte Music Modernization Act. So I think it’s no really surprise that Senator Hatch, he was a key leader on the intellectual property subcommittee, but he was also a composer and singer, a country western composer and singer. You may or may not know his song unspoken. It doesn’t have quite the ring of some of the Taylor Swift, but he did get platinum on that song, and alas, he passed in 2022 and Congress has changed in recent years. Since 2022 we’ve had 44 copyright bills introduced, and they have all gone nowhere. So I think it’s actually a good starting point to talk about how the Act has fared as the courts began to interpret it, and what happens when you have these legislative aspirations either collide or meet with judicial application. Now the question for me is not whether the ACT got the future right, but the more interesting and interpretive question is, was the statute equipped to absorb what we’ve seen and whether we’re actually now reaching some of the limits of what the statute can do. So today, I’d like to address just two major technological changes with you, that is the internet and artificial intelligence. And I think that the act has done quite a good job of absorbing these technologies. So when it comes to the internet, the story of copyright is one of scale, and really we all know the internet fundamentally changed the way that a work would circulate. It was decentralized, it was cheap, and it was global. And so these old questions of copying and distribution, they didn’t just disappear, they just took on a new valence. So courts had to adapt to that scale. And so before we knew anything about platform architecture or user generated content, in 1993 the courts confronted the very first case of internet and copyright, and it was called Playboy versus Frena. Now, what I’d like to show here, though is that at this point, the internet is just inching into the public mainstream, 6.3 million users today, 6 billion. Now the Playboy case involved a bulletin board. It was quaintly accessed by telephone modem. Some of you might be senior enough to remember when you logged in and you would wait and wait and wait. Well anyway, on this amazing modem that they had, you could get 170 Playboy images. So it wasn’t really rocket science for the court to say that is copyright infringement, but it’s also possible to read that opinion as saying, really, copyright infringement is strict liability, and the Court said that either intent or knowledge just doesn’t matter. Now, around the same time here in San Francisco, we had a court that decided a case called Religious Technology Center versus Netcom, and it’s, again, it’s a bulletin board case, and Netcom allowed users to upload a whole raft of information that had been put out by the Church of Scientology. The court there concluded, however, that Netcom could not be directly liable. They just made the bulletin board available and it did not cause copying. So now we have two decisions pointing in opposite directions, and they set the stage for what is this continuing dance between the courts and Congress, and Congress did step in with the DMCA in 1998 and in doing that, if you happen to like legislative history, you would find that they actually cited both of those cases and talked about some of the background. Now, the problems that were presented by Frena and Netcom, they were really pretty easy, but the harder cases came with the reality of peer to peer distribution services. In other words, what happens when the infringer supplies a global infrastructure by which others can circulate copyright content? So in that regard, some of you may remember 2001 the Ninth Circuit held that music sharing service could be liable for contributory and vicarious infringement, it was a facilitator, and the court found no fair use. Now interestingly, that was a time when my son also was becoming an internet user, and he was apparently doing artwork in school, and I think we must have talked about copyright a little too much, because he put a little C with a circle around it on his artwork, and the teacher said, What’s that? And he said, it’s a copyright notice. So anyway, he was well steeped in the world of copyright, but it’s interesting that the court there did reference legislative history. So let me make a little short diversion here, because this question of legislative history has come up in several of the panels. I don’t think it’s fair to say that we have no rules of interpretation. We don’t have what might be considered the 10 Commandments. We have Justice Scalia’s big fat book on interpretation, and we have all kinds of canons of interpretation. But I think what’s interesting is that even though now there’s been some move toward textualism or originalism, and therefore it triumphs over legislative history. That’s not necessarily true, so I will point you to a recent blog or article by Abbe Gluck of Yale, in which she says, actually, legislative history lives on, but in secret. And the way this happens is that current justices and judges cite a case which is precedent, and if you do read the case, you’ll see that that case cites legislative history. So instead of saying up front, I’m citing legislative history, it’s been done a little bit under the table, if you will. So I think that not acknowledging the importance of legislative history. It’s really just to put blinders on. Our job, of course, is to take the text and we have to interpret it. So judges often do that in different ways. They’ll say, here’s the text, here’s what it means in our interpretation, and here’s what it means today in the context of this case, but even so, you can look at the legislative history, because that confirms or is consistent with what we’re doing. So I would say we have not thrown out legislative history with the bath water, and it still has some value. Now, apart from Napster, we then had Grokster 2005 and that pushed the doctrine even further in terms of software liability for inducing infringement. And now, of course, we have more recently, the Cox case of Cox versus Sony Entertainment. There, the court overturned a $1 billion jury verdict, and the Court said that Cox wasn’t liable for willful infringement just because it knew that users of its service were infringing and didn’t terminate their access. So the decision here, I think, froze the vicarious liability doctrine, but it also left a number of questions unanswered. But in these cases, the Supreme Court stepped in to resolve lower court either inconsistencies or conflicts. So taken together, I think these cases suggest that the internet’s first challenge to copyright lay really in the scale at which infringement was occurring, and they also suggest that the primary challenge in this context is to identify where liability should lie in an environment where there’s more than one way to be an intermediary. So I have this cartoon here, and I’ll tell you a cute story about it. I kind of like this cartoon. And I wrote to the illustrator and said, hi, I’m going to give a copyright lecture. Could I use your cartoon? And he wrote back and said, and because I always ask permission in these cases, and he wrote back and said, well, actually, that’s how I make my living. So would you pay a $50 license fee? But if you don’t want to do that, would you just send me a book of your choice? So I sent him $50 of course, and I sent him the book that I had written on Justice Douglas, and now we’re in an interchange, right? But you know, after Grokster, Napster, et cetera, Congress really didn’t step in. There were efforts. Senator Hatch, back to Senator Hatch, he wanted to define what is intentional inducement. But so the years following these cases, what happened was exactly this, Congress did focus on expanding criminalization. So what do you in for copyright infringement? Of course, we now had increasing criminalization through the No Electronic Theft Act, the Digital Theft Act and the Anti Counterfeiting Act. Now, 16 years ago, we celebrated the 300th anniversary of the statute of May, and it was a gathering much like this, sponsored by Berkeley. And I remarked then that Napster and Brungster and their progeny would probably just be a footnote in some treatise or law review article, so those debates are no longer front and center. But I think, as Cox demonstrates, it doesn’t mean that the issues have actually gone away. But those cases, I think in the last 16 years, some of the more interesting questions have been the intersection of free speech and copyright. Someone may have mentioned the Ninth Circuit case in Garcia V Google, and I think it’s illustrative. So there, Cindy Garcia appeared for five seconds, and she had two lines in a film that was supposed to be an action thriller. Instead, it was an anti Muslim film, and it was translated into Arabic, went out over the internet and caused all kinds of violent outrage and protest in the Middle East. And for good reason, so she tried to obtain relief first in the state court, saying it was hate speech. She had violated her right to privacy. I might say no one could ever find a contract that related to this. So that’s one of the odd things about the case. But after her state law claims, she came to federal court, and a panel of our court divided gave her a preliminary injunction on copyright grounds, and then YouTube was required to take down the film. So we heard that case en banc, and we dissolved the injunction. Why? Because she couldn’t really assert what was a copyright harm, and her five second performance was not a work within the meaning of the Copyright Act. So there was despite what a despicable situation it was, and indeed she had suffered, copyright was not her remedy, despite her free speech claims. And so we also said basically that a weak copyright claim could not hijack the First Amendment. And this was an era when we began to see a lot of these cases of hate speech, copyright and speech claims, including from Hulk Hogan, and that era has now passed. But what Garcia V Google did highlight, of course, is this continuing and pre existing tension between copyright and the First Amendment. But as we all know, the Supreme Court has been consistent in its view. As the Court wrote, The Copyright clause in the First Amendment were adopted close in time, and this proximity indicates that, in the framers view, copyrights limited monopolies are compatible with free speech principles. But another principle that I think Garcia highlights is that in the internet world, in the global circulation, it’s not pegged to jurisdictional boundaries. Copyright, of course, is a national regime, and we’ve had discussion already today about the number of international conventions that we have joined, but remedies are issued by particular courts in particular cases and against particular defendants. But the story, I think, of Garcia, illustrates how this internet and copyright really does transcend national borders. And so courts, because we are domestic courts, we still operate in a digital world that is both territorialized and globalized. And as you can imagine, enforcement of copyright beyond national borders is fraught with in court. Incredibly practical and legal difficulties, and it’s a challenge that we continually grapple with because of the cases that are brought before us. So let me turn to a second technology revolution, and that’s AI, so there’s little doubt that generative AI, it’s a textbook example of disruptive innovation, as Clayton Christensen from Harvard put it. And I kind of like this quote from Asimov, because he describes AI or any of this kind of technology as kind of magic. And I suggested that to Brad Smith of Microsoft when we were on a panel, and he said, no, no, I disagree. It’s just an algorithm. So it might be just an algorithm, but it’s causing conniptions in the copyright world. And of course, I just want to point out this is the evolution on authorship of the different words we’ve used in the Constitution, the 1909 Act, and, of course, the 1976 Act. And I think that what the court what the legislative history shows in the 76 Act, is that original works of authorship, it was actually purposefully left undefined, so that it now falls to the courts to define and implement. What does that mean? And I can’t really leave this talk without my favorite or one of my favorite Ninth Circuit cases, and this is our friend, Naruto the monkey, and we held in 2018 that the monkey lacks statutory standing to pursue a copyright claim, so we can laugh and whatever. But in fact, Naruto was the beginning then of copyright guidance at a copyright office, and they used this case to explain the contours of human authorship and the human authorship requirement. So it may be that this notion of human authorship is more load bearing than we originally thought. So what kind of human create, creative agency is copyright trying to honor, preserve or privilege, and as relevant to fair use, what kind of markets is copyright imagining? And so while the assumption of human authorship is one of the key and fundamental pressure points that the act is facing, I think it doctrinally, harder is the question of fair use. Of course, easy to talk about. You can cite the four factors more difficult to apply. The Supreme Court has added texture in at least three key cases, 1985 Harper and Row, the court said, and we still repeat again and again that the fourth factor, market effects, is tagged as the most important. Then we fast forward to Google, the Oracle. 2021, that’s a software case, and it was Google’s re implementation of Oracle’s declaring code. It was transformative. According to the Court, Justice Breyer there reiterated that fair use is an equitable rule of reason, and it permits courts to avoid rigid application of the copyright statute went on occasion, it would stifle creativity, which that law is designed to foster. And then, of course, we come to the Andy Warhol case 2023 and even though Andy Warhol’s use of these photographs of prints added some new expression, there was no fair use stopping on the first factor. So now, how will these principles play out in the future? It’s interesting. We have now about 80 generative AI cases in the courts, and most of them are class action suits, not all, but you can see from the slide that they kind of segment in to medium, whether it’s text, music, computer code or whatever. But as yet, I might remind you there’s no appellate decision. I will talk briefly about Bartz V Anthropic. It’s easiest to talk about because it’s settled a whopping $1.5 billion settlement. No appeal, of course. So there what Anthropic did, It downloaded and digitized millions of copyrighted books from pirate sites on the Internet and some from copyrighted books that it had purchased. And the goal is they were going to make like a central library of all the books in the world, and then use it to train Claude, the decision, I think, is interesting, because it illustrates the challenges the courts have. And instead of saying all training is transformative and fair use, what Judge Alsup did is he put it in buckets, and he said it depended on whether they had legally purchased and digitized the books that was justified as fair use, but the pirated books was not. But I think the harder question that the opinion grapples with is just, how far can you push the first fair use factor in transformative use and the fourth factor on market impacts. I thought it was interesting. He characterized generative AI as quintessentially transformative, and he described the technology as probably the among the most transformative that many of us will see in our lifetimes. So it’s interesting. He also said, Well, it’s just training on it is like a reader aspiring to be a writer. Readers study text and they learn how to create original works. So LLM’s read text and then they produce fundamentally different works. And what you see here is what we often see in court decisions when we particularly in technology, we try to analogize to something we know and understanding it does fit, sometimes, sometimes it doesn’t fit, but we’re often reasoning by analogy. So for lawyers, academics and others, I think it’s very important to peg some of these technology cases in analogies that the court can understand, Anthropic of course, said, well, the training is fair use because the existence of a pirated library shouldn’t even matter. Judge Alsup projected that argument, and he said, there’s no kind of free floating entitlement to copy first and justify later, so we can debate whether that decision was right or wrong, but I think it’s fair to say that one thing it did, it rejected AI exceptionalism in the world of copyright. It’s also notable for what it didn’t do. He rejected the idea that generative AI will flood the market with competing books. Now, some might and have disagreed, so I’ll mention briefly the case of Kadrey V Meta, which is still pending in the District Court. But there is a district court opinion on a summary judgment, and Judge Choudhury has said, well, the use of these books from various libraries like Genesis and Anna’s Archives, it was fair use on the record before him, but he also quickly highlighted that this holding didn’t actually establish the general legality of LLM training, because, in his view, the LLM training combines two features that we hadn’t previously seen. One, a deeply transformative use of the work, and two, the potential to flood the markets. So again, the first and fourth factors become the most important. These cases suggest that the battleground going forward really will be on what kind of market impact is the copyright law prepared to recognize? And I want to emphasize that these are only just a handful of cases. Even these are just cases in which there’s been decisions, but there are scores out there that are still pending with not even a district court decision, and there’s no appellate decision. So I wonder, is this like the early software cases on copyright, for example, the 1986 case of Whelan V Jaslow Dental, that talked about structure and sequence and organization of code, and then was later disfavored and basically jettisoned, and it was one of the early cases. So the question remains, are these early cases that will be adapted and adopted, or will they be jettisoned? I think also the cases suggest a somewhat different way of describing the current moment? I predict in the case of AI that the courts will be first movers and not Congress, and Congress will be a more leisurely follower. And so these cases are really understood as disruptive innovations, and as such, I think it’s an overstatement to say that AI is a rupture of copyright history rather than a continuation of copyright history. But as we know, the courts can only go so far. And so the solution some have proposed is something systematic. Let’s have a compulsory and collective licensing regime. So that proposal has attracted as many proponents as critics, but it’s the kind of question that’s for Congress, not for the courts, and Congress is well equipped to answer it, but it might not so again, a legislative act going back to Justice Holmes both look. At the future and sets out rules of conduct that apply to more than just a few people. So the courts cannot rewrite the statute and write on a broad slate, but I think Congress likely will not. So Congress might not be moving, but there’s a few other institutional players to look at. The states have been aggressive, and there’s a whole segment of a conference on the states. All 50 states have introduced AI bills. 38 of them have enacted some that are some kind of a specific AI measure. You know, Colorado, for example, talks about algorithm discrimination. Utah ties its case to consumer protection. We know in California there’s been transparency rules. So these laws may not directly relate to copyright, but I think they that remains to be challenged, and I think you’ll see preemption arguments on the horizon, but these state laws may well reflect the laws development, and they can force disclosure about training, for example, or create evidence about developer practices. And I think that they will shape copyright litigation, records and evidence in ways that are yet to be determined. So there we have the states, a new player, and at the federal level, we have basically executive orders. You might be interested to know that President Trump has issued 475 executive orders over his two terms, but the one relevant here, Executive Order, 14179, put that to your memory. That was issued in 2025 and it really pushed toward a minimally burdensome national framework. And it said, of course, we should have legislative proposals for uniform federal policy, and that’s unlikely to happen, at least in the near term, and it targeted these state AI laws and viewed them as obstructive to innovation. So now we have the federal government stepping in the States, but we can’t forget Europe, because Europe’s framework for copyright claims and generative AI, it’s well past the early stages. It was an early mover in this area. We had a 2019 copyright directive and related rights in the digital single market, and there they talked about copyright exceptions for text and data mining. Then 2024 we layered on transparency, safety and other obligations that relate to rights holders, opt outs and by requiring general purpose model providers to summarize the content used to train their models, the AI act in Europe has given that European copyright law ex anti regulatory and compliance dimension. And because the market is global, Europe is a critical player. So the dance has really expanded well beyond the courts and Congress to the states, the executive and Europe and recall in 1976 these other institutional actors, the states, the executive orders and the European dimension were not on the drawing board. So what all this suggests, I think, is that we’re living through a period when the statute becomes interesting in new ways. We have the acts, words, originality, authorship, work, copy and market, and they’re all asked, being asked to do new work in a different factual circumstances that were previously truly unimagined by the drafters. So what are we going to do? Well, courts are going to do what we’ve always done. We need to interpret these definitions somewhat flexibly, because we have to accommodate the innovations, and we can’t turn every technical shift into a legislative emergency. So we are watching copyright in real time now, and we’re giving new meaning to some of these words, and I think that actually is some evidence of the acts success 50 years on, it continues to give us language to debate and argue about and talk about what comes next. You all recall those various Supreme Court pronouncements that courts are not well situated to be art critics. Well, I would say as to AI, judges have no choice but to face it head on and to become interpreters of AI. So I leave this challenge especially to the law student. Is here because I don’t expect to be here for the 75th or the 100th anniversary of the 1976 act or its follow on, but I would say you will be. And so I thank you for having me today, and we’ll take some questions.
Erik Stallman 35:24
So, judgeMcKeown has graciously agreed to stick around and answer some questions. There will be people coming around with microphones. Please, do raise your hand if you have a question.
Audience 35:42
Thank you so much for speaking. I was wondering what you thought of Judge Alsup comparison to human readers and writers, especially in regards to the arguments that human readers don’t create like transitive copies as they’re learning, and that’s potentially a concern for AI training. What do you think of those analogies and the assumption that AI is inherently transformative.
Hon. M. Margaret McKeown 36:02
Well, okay, let me first say these are the kind of questions I have to take the fifth on. And why is that? Because the same arguments are actually percolating through the courts now. So I kind of leave to you. I mean, we use analogies all the time when we had other cases, we talked about, oh, it’s just like a copy machine, and somebody uses a copy machine at your library, we’re not going to hold the library responsible, although you all see those copyright notices over the machine put there by the General Counsel’s Office. But I think what it tells you is that what Judge Alsup was trying to do, is he was trying to bring it into a way that we could understand, in a framework that we already have now. Was it a perfect analogy? Is it the right analogy? Luckily, we don’t have to pass on that, because since it’s settled, we don’t have the opportunity in the ninth circuit to take a look at it.
Erik Stallman 37:04
Well, while we’re waiting, just one question I’d like to ask, when you put up sort of the slide with all of the cases, which aren’t even all of the cases. So one could despair, because it doesn’t seem like at any point soon we’ll have a sort of a case that resolves all of the questions that we need to decide, the questions of copyrightability and copyright infringement relates to artificial intelligence, and some of these cases are actually quite narrow. For example, Doe versus GitHub is deciding on an interlocutory appeal only the question of whether or not the accused infringing work must be identical to the original before you have a claim for removal of copyright management information. So from the perspective of a judge who’s deciding these cases, are you, are you comfortable in a world where you’re living with a lot of atomized precedent you’ve had to sort of piece together yourself? Or do you find yourself sort of wishing that, if not Congress, and likely not going to be Congress, that someone would piece this together for you, so there was more of like a comprehensive authority on these questions.
Hon. M. Margaret McKeown 38:06
It’s a good question. I might say the GitHub case has been argued in the ninth circuit, so there will be at least a decision on that. But you know, that’s the world that we operate in. We are deciding individual cases on individual facts. And while in an ideal world, you know, we would put them all together and create some pastiche of copyright and generative AI, that’s just not the way the courts work. And I think part of the really, the beauty of these cases coming in different kind of iterations at different phases is that we are going to begin to see some copyright principles that can then be applied to others. But what we’re doing now is we’re not making up the copyright law. We’re not jettisoning the 1976 Act and the 79 amendments to that act, but we’re taking prior precedent, and we’re basically applying that. So I don’t find it to be too frustrating. I think it could be frustrating to academics and practitioners. You have a case, oh, this case is a little different than that case. Well, I love the advocacy system in the United States because I think that it really brings to fore the ability to give a statute some life. You know, it’s 1976 we got some really beautiful descriptions of everything that went into that. But that’s 50 years ago, and we’ve had a lot of technological developments, you know, we, we, of course, I mentioned to the internet, but we, you know, we had cable television, we had a whole series of technology related cases, software that came before these. And so, you know, for the courts, this is not something unusual, and we don’t find it particularly frustrating that we’re asked to decide a particular case, now, then we get into the question of what’s dicta. So if you write too far, either your colleagues or another court will say, well, that’s dicta. You don’t even have to listen to that. And the Supreme Court has some interesting thoughts on dicta. But again, I think that’s kind of the bread and butter of what we do, and at this early stage, it might feel frustrating, but for those of you who are here in the copyright of software era, and I was, you know, a young practitioner at the time trying to register various software and litigating cases on software, that was the situation we were in. We didn’t know quite what was the framework or the parameters of software and copyright at that point, you know, despite CONTU And despite the enactment of the 1980 software act. But you know, it wasn’t long before we actually had a baseline and we had principles, and we were applying those. So I don’t think that generative AI will be any different, but I do think it is a kind of cataclysmic technological innovation that does surpass some of the other innovations that we’ve tried to adapt to.
Audience 41:29
Hello, I want to ask a question about amicus briefs, because with all of these cases pending, I have a feeling a lot of people in this room are going to be participating as amici. And so I’m curious, from your perspective, what do you find most useful and effective from from amici, and what would annoys you? And what would you say, like, please don’t do this in your briefs.
Hon. M. Margaret McKeown 41:53
It’s a great question. Of course, we’re often inundated with these amicus briefs. And the one thing I will say, I don’t know that it irritates me, but I’ll say it’s not helpful, if you only have a me too brief. You know, when we think about these, what I’ll call transformative amicus briefs, we think to Brown versus the Board of Education, and of course, they were using social science that was not part of the record in those briefs to talk about the impact of the separate but equal situation in the schools. So the other thing I think we have to be careful about, though, is to recognize that we are deciding a case on the record, and you have to be incredibly careful not to buy the Kool Aid of an amicus brief that sounds beautiful, but it’s so far outside the record that you’re deciding a case that wasn’t in front of you. And I think you all might remember that Supreme Court case, Bush versus Gore, and when they dug into it, one of the things the Supreme Court cited was like a news article from some newspaper in Iowa. I think it was, and it was written by what people called the broom closet guy, because he wasn’t even a reporter, and he didn’t really understand some of the significant principles in studies that have been done on elections. So they actually cite this, and it was totally wrong. So that’s what we have to guard against. But what’s helpful in an amicus brief? Well, I think it’s helpful to explain the policy ramifications, because when you are interpreting a statute and a section of the statute in your head, you’re wondering, how is this going to run through the rest of the statute? And we’ve seen so many sections up here of this and that and the other so that’s very helpful to us. And again, yeah, of course, that’s policy, and we’re not going to make policy, but I think it helps us understand, does it help to have 100 law professors versus 20? I don’t think so, but it is. It’s quite impressive that you can get 100 people to sign on to anything, actually, particularly law professors. So the kind of amici we get tend to be law professors industry. And then you have to be very cautious about, well, what segment of industry, you know, what might be the motivation behind this? But I read those amicus briefs, and I think they’re quite helpful now the Supreme Court. Now you don’t even need permission. You can file an amicus brief, and they’ve got, you know, we do it all digitally, but you know, to the ceiling amicus briefs, and you have to say to yourself, well, what really would be helpful, in addition to pointing out ramifications across the statute, what the policy issues are, what some of the pitfalls, I think it’s also possible to take arguments not made by the parties in. But that might be embraced within the party’s arguments. So we have the idea of party participation and party presentation. We’re not just supposed to wildly go off and create issues, but if you have a particular gloss on what could be within an argument made, you could also talk about arguments not made, and why not made, and what those would be. So I find that amicus briefs can be particularly interesting. You know, over many years, most other courts didn’t have amicus briefs, particularly in Europe, and they’ve started to change that policy, and we’ve had a lot of discussions about that, but I think that, you know, we have to recognize where is that brief coming from. Why are they investing in an amicus brief on this subject? So, you know, we take it not with a grain of salt, but we with sort of our eyes wide open. But again, I think we welcome and particularly now another thing that could be useful, I think, is implications of the technology. So you might think you’re deciding this itty bitty little case, but here’s a broader implication of the technology, or you might give us a better explanation of the technology that you can do within the ambit of the case that’s actually being argued. So again, bring on those amicus briefs.
Erik Stallman 46:35
Thank you so much.