Kent J. Schmidt

Transcript

Schmidt

[00:00] Welcome to another episode of SharkCast.  Today I’m pleased to welcome back to the virtual SharkCast studios my partner and friend Mike Keyes who’s back for another episode following his first episode a number of months ago.  So, Mike welcome back.  Thank you for rejoining us here at SharkCast.

Keyes

[00:22] Kent, great to see you.  Thanks for having me back.  Always a pleasure to be with you.

Schmidt

[00:26] Well Mike, today we’re gonna talk about AI and copyright.  And before we dig into copyright law, just an observation, it seems like in almost every conversation with family members or colleagues or neighbors or whatever, somehow, someway, AI comes up.  Because everyone’s starting to discover the millions of applications of AI and it’s also the fact that AI is covered in conferences and events and CLEs continually and we want to make we’re adding new value added material and not just mimicking the content of others.  But do you find that in your circles that AI is just sort of a topic of conversation everywhere you turn.

Keyes

[01:18] I think that’s exactly right Kent.  And at the risk of overstating it, I think it’s probably the biggest copyright related issue in our history frankly.

Schmidt

[01:28] It is, and it’s just permeating in our lives.  I chuckle when I think about this, but I decided a couple weekends ago to try my hand at making homemade French baguettes with flour sourced from France.  And I didn’t use a cookbook, I used a combination of YouTube videos and AI to help me through the process and to sort of learn the process for the first time.  They turned out okay...

Keyes

[01:58] I love it.

Schmidt

[01:59] ...But not like right up to the standards but.

Keyes

[02:02] You’re not going to quit your day job.

Schmidt

[02:04] No, not quite yet.  But if I’m now using AI to make French baguettes, then I think it can be used for virtually anything.  And of course, no industry has been impacted by AI as much as what we would call the creative industries, you know, music and books and what is, comes under the heading of publishing but perhaps even broader than that.  AI is allowing everyone to be creative, everyone to be a publisher of course.  And there in lies the problem.  So, Mike you come from a background of IP litigation with particular emphasis and focus, at least in part, on copyright, so I’d like to chat with you about copyright infringement claims and how they’re being impacted by AI.  We have a number of topics we’re gonna unfold here but can you just give us a general observation at the outset about this.  And what I like to do is hear your reaction then we can give it our listeners a little bit of a level set or refresher on some AI concepts and then talk about a decision in particular relating to AI and conclude with some thoughts on litigation risk avoidance, my favorite topic, in the context of AI.  So...

Keyes

[03:42] Yeah.

Schmidt

[03:43] ...What are your overall thoughts on AI and copyright infringement in particular.

Keyes

[03:50] Yeah, I think you really hit the nail on the head at the outset of today’s episode.  I mean it’s such a gargantuan issue in the world of copyright.  As I mentioned, I think it’s the biggest copyright related issue that we’ve ever faced in the history of copyright.  And copyright goes back to the founding of our country.  Our first copyright act was enacted in 1790.  It was very limited at the time, as you can imagine, it only protected, I think it was books and maps, at the time, and charts maybe or graphs.  But since then with technological revolutions including the player piano in the early 1900s, to radio broadcasts in the 20s and 30s, to Sony Betamax machine from the 1980s, the digital revolution of course of the 90s and 2000s...

Schmidt

[04:43] The Napster era.

Keyes

[04:43] ...And now of course we’re at AI.  Napster, exactly, right, right, right.  So we’ve seen over the course of our history a number of technologies that have come into play that didn’t exist and copyright had to adapt.  Either through court decision or frankly it was more court decisions trying to catch up with copyright and then Congress stepping in to amend the Federal Copyright Act.  So, I think we’re seeing, we’re on the cusp of a, not on the cusp, we’re in the kinda the bullseye of copyright evolution 7.0 or whatever it would be at this point with AI.

Schmidt

[05:25] Well, and you mentioned that concept of courts trying to catch up with technology and of course that relates to how swiftly technology is changing and we’ve never seen that acceleration quite like we have in the last two to three years when AI has come into I guess common use.  I mean AI has been around much longer than the last two years, but...

Keyes

[05:52] For sure.

Schmidt

[05:52] ...It seemed like it just hit, you know, homes and consumers within the last couple years and so it’s moving, the technology’s moving so fast, and courts are really having to catch up, right.

Keyes

[06:09] Yeah, that’s exactly right.  I mean just like with other technological evolutions, I mentioned the player pianos from the early 1900s where, you know, the supreme case involving piano rolls for player pianos.  It went all to the US Supreme Court.  And the Court said well this really isn’t a copy, you know, the perforated piano rolls that used to stick into player pianos, Court said that’s not a copy.  Well, of course, Congress stepped in and amended the Copyright Act to deal with the notion of what are called mechanical licenses, and we still have mechanical licenses that are part and parcel to the Copyright Act today.

Schmidt

[06:48] So, you’ve laid a little bit of the groundwork but lets, you know, talking about the origins of copyright law, but can you give us just an overview of US copyright law in just a few minutes.  I know that’s a challenge.  With particular emphasis on the fair use doctrine, which a lot of people invoke and use sort of like the phrase hearsay or other legal concepts that people through around but actually has a specific and defined meaning.

Keyes

[07:23] Sure, well I’ll do my best here Kent.  So, as I mentioned our first copyright act is from the 1790s.  Copyright is a concept that’s actually enshrined in the Constitution in Article 1, it’s one of those specific instances mentioned in the Constitution that Congress has the right to protect artists and inventors, or authors and inventors and give them a limited monopoly over their works and inventions.  And so that led to respectively the Copyright Act and the Patent Act.  The Copyright Act has been amended over the course of many years but it essentially protects in its current formulation what are defined as works of authorship.  And you can think of that as being fairly broadly defined to include things such as books, movies, architectural works, paintings, you know, creative works if you will as a shorthand.  And so it gives certain exclusive rights to authors that allow them to prohibit the copying of those works, distribution of copies and other related rights.  So that’s essentially the bundle of sticks that copyright authors enjoy.  Now you mentioned the concept of fair use, there’s also built into this statute this whole notion of fair use, in section 107 of the Copyright Act, that provides that it shall be an affirmative defense to a claim of copyright infringement, that someone has used someone else’s work in a manner that the law considers fair.  Kind of the quintessential fair use would be in the academic realm, it’s not exclusively for academic related purposes but, you know, a professor copies a portion of an article and distributes that copy of the article to his or her students to teach a concept.  We think of that being kind of quintessential type of fair use, but there are certainly other instances too.  We’ve a couple of recent fair use cases over the last, it’s been over the last few decades now, but most recently a couple years ago the Supreme Court weighed in on the whole concept of fair use and copyright.  So we have a bit of guidance from the courts, but ultimately it’s such a fact specific issue, it’s going to turn on the set of facts that are presented to the court in the context of litigation between the parties that are at issue in the lawsuit.

Schmidt

[10:00] And can you comment briefly on registration and how that impacts the rights of the author to assert a claim for copyright infringement.

Keyes

[10:13] Sure, so under the Copyright Act, there is a method by which you can register your copyright with the Library of Congress.  It’s a fairly straight forward process.  You take a copy of your work, you submit it along with an application electronically to the Library of Congress and they examine it, there’s an examiner at the copyright office within the Library of Congress that will take a quick look to make sure that the prerequisites are met and then you’ll get your work registered.  Now you don’t need to register your work in order for you to have protection.  Under US law you receive copyright protection once your work is what we call fixed in a tangible medium of expression.  So, once it’s fixed and it’s perceptible you have copyright protection in that.  Now it’s always best to register your work because you get certain enhanced remedies as a result of registration.  And if it’s a US work you actually can’t sue copyright infringement until it’s registered with the Library of Congress.

Schmidt

[11:18] I see.  Okay, those are important foundations for our discussion that follows.  But before we talk about AI, can you explain some of the ways in which copyright law has been adapting to what we might say a modern technology including software code, including new digital applications and so forth that I think lead up to the courts deciding cases involving AI.

Keyes

[11:54] Sure so, I guess I’d go back to the 1990s when the internet exploded and we had all sorts of issues with distribution and copying, making digital copies and reproductions of works.  That’s one of the instances where Congress stepped in and passed what’s called the Digital Millenium Copyright Act.  And so it’s been with us for a couple of decades.  It works reasonably well.  But we see copyright infringement related claims in all sorts of concepts both in business and non-business setting alike.

Schmidt

[12:13] I mean one way I would understand it is that all that technology has made it easier to copy all types of things, all types of media, and with the ease with which people can copy and transform, which is something we’re gonna talk about in a minute, or transform to at least a certain extent, there are more copyright claims than ever because everyone has this ability to copy that you didn’t have before, that was more tedious to do before.  And so every time there’s a technology that allows someone to copy you have more and more copyright claims.  Let’s talk about the transformative aspects of a copyright claim.  Just putting it in very basic terms, how courts have dealt with this concept of the ability of a person to take something that has protection and then transform it until it no longer is a copy, but it is something that’s new.  What are the general principals that relate to that.

Keyes

[13:44] Yeah, so we could probably have a whole series of podcasts related to transformation when it comes to copyright.  But essentially, I think it’s best to think of this in, with the kinda dual principles associated with the Copyright Act.  What we’re trying to do is, we’re trying to incentivize authors and other creators to create works of authorship, to have a really robust output of creative content.  Yet at the same time we’re also trying to balance that people will take another work and add to it and transform it and add to the betterment of society as a result.  And so we have on the one hand we want to provide exclusive rights to authors so that they make sure that they’re incentivized to create works, but at the same time we don’t want to squelch the creativity completely of other people.  We want them to be able to, you know, we all stand on the shoulders of giants, right.  So, we all want to incentivize future creators and authors to take what’s in the existing corpus and be able to transform it and use it to create more works.  So that’s kinda the inherent tension that we have between copyright protection on the one hand and fair use on the other.  And one of the elements that courts really look to try and balance those somewhat competitive interests is to look at, and this is one the probably the most important fair use factor, it’s the transformation that has been where the second comer has used the first creators work and transformed it in a meaningful way to create a new purpose or to create a new work that still incorporates parts of the preexisting work.

Schmidt

[15:44] That inherently sounds like a very difficult line to draw just by your description and it’s a great segway to talk about the case that we’ve teed up here.  It is Andrea Bartz et. al. vs. Anthropic, the AI, one the major AI platforms.  It was issued by Judge Alsup in the Northern district of California in June of 2025.  So why don’t we talk about this case, this important case.  I’ve enjoyed reading it.  We’ve had some discussion back and forth offline, but for our listeners can you summarize what the case is about and then we’ll get into how Judge Alsup applied some of these principles we’ve been talking about in the brave new world of AI.

Keyes

[16:36] You bet.  So this is an actual class action lawsuit that has been filed by a number of authors.  And as you mentioned they sued Anthropic which was actually founded by a number of former OpenAI employees that launched this new venture back, I think it was in 2021.  And so they launched this new venture Anthropic which is a large language model that’s branded under the name Claude.  And so what they did here, allegedly according to the lawsuit, is they did a couple of things.  They trained Claude using I guess two different types of works although they were all works of authorship, they’re all books for the most part.  What they did initially is they obtained what Judge Alsup refers to as pirated copies of some of these books that were available on the dark web and other locations where you could actually go and download them.  And so, Anthropic initially started using pirated copies of books and a huge number of them to train Claude, but then ultimately started using actual copies that they obtained legally.  And what they did, they obtained copies, Anthropic did, and its employees and other agents, obtained copies of, physical copies, of the books at issue and then digitized them to create a massive library and also used that to train the LLMs.  So the authors were not happy about this, claimed that the process of taking the works both the properly obtained physical copies and digitizing them was an act of infringement and then also using the pirated copies to train the LLM was also an act of infringement.  So that’s the basic claim at issue in the suit.

Schmidt

[18:44] And the decision that we’re going to discus was on a motion for summary judgment which...

Keyes

[18:50] Right.

Schmidt

[18:50] ...It was interesting to me just as a litigator that they sort of bypass class certification and bypass a motion to dismiss and just went straight to summary judgment and front and center in the motion for summary judgment is whether this is a fair use, of course a distinction being drawn between the pirated copies and the copies that were legitimately obtained.  So how did Judge Alsup come down on these issues for training the LLM with both pirated and legitimately obtained copies.

Keyes

[19:27] Yeah, so we talked a moment ago about the whole notion of transformative uses and conceptually what Judge Alsup said was, look, for purposes of training the LLM, this really is a very transformative use of someone else’s copyrighted work.  I think he said, exceedingly, it’s exceedingly transformative.  And quote, spectacularly so.  So Judge Alsup really is onboard with the whole notion of, at least in theory, using someone else’s copyrighted work to train an LLM is very transformative in it’s own right.  So, in that regard, certainly that was welcome news to Anthropic and other AI developers, that look, we do have a pretty solidly in our camp ruling that using copyrighted works to train an LLM is in fact transformative and is a fair use.

Schmidt

[20:29] Let me interject a question just to make sure I’m clear as from a very practical standpoint what you’re understanding is as to what we’re talking about for training.  So for example, I think my son’s favorite, one of his favorite authors is Cormac McCarthy who writes in a very unique style and so if you uploaded all of Cormac McCarthy’s novels, however many there are, I think he passed away a year or so ago, you could then allow a user to say I want you to write this in a Cormac McCarthy style, a subject completely different and it would understand that style.  Is that a correct understanding as far as you know from what is being referred to here as training the LLM.

Keyes

[21:25] I think that’s right.  And there’s a very important limitation on what this case is about.  This case is about inputs not about outputs.  So, what I mean by that is all Judge Alsup’s ruling deals with is whether the concept of training an LLM with copyrighted content, whether that’s a fair use.  And we do have an answer on that rather narrow but very important issue.  There’s a whole other outstanding issue as to, well what about the outputs, what are those, are they potentially infringing.  It doesn’t rule on that specific issue at all, doesn’t even touch it.  And it says look this is just about, Judge Alsup says this is just about the whole notion of training.  So, we only have really one, an answer to part of the equation here when it comes to a claim for copyright infringement.

Schmidt

[22:25] So lets extend the Cormac McCarthy illustration to see if we can understand the inputs and outputs.  The scenario I just described where the, you know, let’s assume that they’re all legitimately obtained, someone goes buys all the books, a copy of all the books, and they train an LLM to write in the style, or put out in the style of Cormac McCarthy, that’s be a legitimate output, legitimate input under the, Judge Alsup’s formulation but a question he didn’t decide is that would allow the user to essentially say give me the cliff notes version of the book or give me a condensed version of the book.  That’s an output that he didn’t reach and that perhaps other courts have reached or will reach.  Is that the distinction between input and output.

Keyes

[23:27] I think that’s exactly correct.  He didn’t reach that other very important issue as to the outputs, the content that’s created by the AI.  And whether that would be infringing and who would be the infringer in that instance.  So that’s not answered by Anthropic.  There’s another case, I mean there’s probably I would say three dozen AI cases that are pending in various federal courts throughout the country, but one that was recently filed that I think will get to that issue is a case involving Disney where it sued, and I think there’s both the inputs and outputs issue involved there with respect to that AI.  That was a case that was just filed, I think it’s in, down in your neck of the woods Kent, in central district of California.  And so we’ll see, I suspect we may get an answer both on the inputs and outputs question there.

Schmidt

[24:26] Okay, well I sense another episode of SharkCast maybe a year or so from now.

Keyes

[24:32] There you go.

Schmidt

[24:33] But in the meantime, let me just sort of get, put you on the spot Mike, if you don’t mind.

Keyes

[24:38] Please.

Schmidt

[24:38] And ask you if you’d put the black robe on for a moment, and you’re a district court judge, you’re also a creative guy.  We’ve talked about in the last podcast about your music background, you have a lot of works that you’ve done and created particularly in the music space.  What are your thoughts on this, on how Judge Alsup decided at least this case and to the extent you have thoughts on these issues that remain to be decided.  Do you think the courts are find the right balance in that tension you talked about between encouraging new works and protecting the authors and originators of works.

Keyes

[25:27] Yeah, you know, I must say I’ve kind of gone back and forth a little bit, but I have a lot of respect for Judge Alsup.  He’s been involved as you know in a ton of cases involving technology.  His opinions are always a good read.  I think his analysis is correct.  I think that in this particular circumstance with the books that were legally obtained and using them for a very transformative purpose of training an LLM, I think that is the type of transformative use that the Copyright Act should encourage.  Now that being said, I think ultimately, we’re going to need some sort of congressional guidance here in terms of what exactly are the proper boundaries here.  You know there is a bill that’s pending, I’m not sure how far along it is, the TRAIN Act.  It was proposed by a group of senators about transparency in training LLMs.  Yeah, I see some utility in that but it only answers part of the equation again.  Or only provides specific information and I’d say the TRAIN Act itself presupposes that training an LLM would be an act of infringement which is kind of contrary to what Judge Alsup just told us, at least in the limited context of that case.  So, I think we’re gonna need to have some sort of congressional action here that really does tell us what the proper balance is.  And how that typically works in the copyright context is you get a whole bunch of stakeholders together, there’s a whole series of hearings and meeting with Congress and they try to hammer out what the statute should look like as a result of hearing from all these stakeholders.

Schmidt

[27:26] Yeah that seems to be a very, very important piece of the process rather than Congress just starting to tackle these questions on its own without that input from the stakeholders.  Well, one of the things I want to cover also because the whole purpose of this podcast and what I write and speak about is litigation avoidance.  So, whether is relates to AI or not, I’d like you to address from a very practical standpoint the steps that companies should take to minimize copyright infringement risk, litigation risk.  And if you can address some of the way perhaps outside the sort of stereotypical basic copyright infringement where someone just photocopies an entire book or something of that nature.  You know stands at the photocopier for hours and hours, that’s kind of an obvious thing.  But what are the ways that copyright claims sneak up and arise for a company that doesn’t intend to violate the law or anyone’s rights and yet they find themselves embroiled in a major piece of copyright litigation.

Keyes

[28:42] Sure, yeah, I mean the number of, or types of instances of copyright infringement are vast these days particularly with AI.  I mean we’ve seen claims related to an employee taking, for example, a competitor’s marketing collateral, feeding it into an LLM and having a whole different type of work created but it has a similar look and feel, it has similar graphical elements and so on and so forth.  Something like that is so easy to do from an employment perspective as far as the employee doing that.  I mean we see a lot of instances where the inputs are provided to the LLM and you ask the LLM, hey, create this output that looks something like this.  I think there’s numerous instances where that sort of thing is happening and can create potential exposure.  So your point about litigation risks and mitigating those I think it’s so important in this context to make sure that, you know, companies should have AI policies related to what employees can and can’t do with respect to AI.  I mean super valuable tool to have at your disposal for doing all sorts of things quickly and efficiently but again need to be careful to balance against the rights of third parties and make sure that you’re kind of staying on that side of the line that wouldn’t lead to a claim of infringement.

Schmidt

[30:20] Well that sounds like a very, very tricky area to navigate.

Keyes

[30:23] Very.

Schmidt

[30:24] Constantly evolving.  And the ability to, as I said earlier, the ability copy means the ability to create a copyright infringement claim so easily.  And I appreciate what you said about policies.  I guess I would add every time I hear about company policies, I always want to add the addendum to training as well.  Because policies...

Keyes

[30:47] So true.

Schmidt

[30:48] ...Policies without training are often, perhaps even worse than no policy at all.  Because you know people violating a policy but not really understanding the policy or forgetting the policy.

Keyes

[31:00] That’s absolutely true.  I mean it’s critical that employees understand at least at a high level what the issues are here and what we’re trying to avoid as a company in terms of running afoul of the law and running into potential copyright infringement claims.  So you’re exactly right.  Policy without training, and regular training and reminders isn’t much of a policy at all.

Schmidt

[31:24] Do you do some of that training in your practice of advising companies on not only their policy but offering some guidance on the training or doing the training yourself.

Keyes

[31:35] We absolutely had.  Particularly in the context of creating advertising campaigns, so yes it’s something that we have been regularly involved in recently.

Schmidt

[31:46] Very good.  Well I like all that litigation avoidance work.

Keyes

[31:50] I know you do Kent.

Schmidt

[31:54] The ounce of prevention is worth the pound of cure in [UNINTELLIGIBLE]...

Keyes

[31:57] Yeah, no doubt.

Schmidt

[31:58] ...Different areas.  Well that is about the extent of our time to talk about copyright infringement.  And as a veteran of the SharkCast podcast you know we like to end with sort of a lighter note, talking about things outside the law, the world that exists out there.  And when we were trying to schedule a time to record this episode you mentioned in passing that you had a road trip and it piqued my interest and I don’t really know much about it.  I know a couple of facts but can you tell our listeners about your recent, sounds like epic road trip that you went on.

Keyes

[32:37] It was rather epic I would say.  So yes, when we tried to schedule this before I said, you know I’ve had something come up.  I’m going on a bit of a unexpected road trip with my son.  We drove from Washington state with a U-Haul and his car behind the U-Haul, drove up all the way up to Anchorage.  2,515 miles to be exact.  So we had an amazing time.  That’s a part of the world that I hadn’t from that perspective.  I’ve flown to Anchorage many, many times but never witnessed it on the ground and it was truly, truly sensational.  Great father, son bonding experience.

Schmidt

[33:16] How many days did it take you.

Keyes

[33:18] It took us five, which was about 500 miles a day, which doesn’t sound that grueling, but when you’re on the Alaska highway and it’s two lanes and you gotta be constantly on the lookout for wildlife and so on and so forth, it’s a long 500 miles a day.

Schmidt

[33:37] Yeah, I’m sure, I’m sure.  A lot of really fantastic conversations with your son I’m sure.

Keyes

[33:44] Amazing.  Yeah, it was great.  It was super memorable.  Now that said I probably won’t ever do it again.  But it was sure a blast.

Schmidt

[33:53] Yeah.  Did you stop in on the Dorsey Anchorage office while you were there do you at the...

Keyes

[34:00] I did.

Schmidt

[34:00] ...to Seattle.

Keyes

[34:03] No, I stopped by and said hi to all of our friends up there and regaled them with some of my Alaska highway stories which people really got a kick out of.

Schmidt

[34:12] Well that’s fantastic.  Well Anchorage, I have not been to Anchorage either by road or flight and so it’s definitely on my list of places to get to whether for business or pleasure or actually probably more likely a combination of both.

Keyes

[34:26] There you go.

Schmidt

[34:27] Thanks for sharing that with us.  Well, that’s about all the time we have for this episode of SharkCast.  I’d like to thank you again for being here and very interesting conversation.  I’m sure we’ll reconnect probably next year to talk about some of the emerging decisions in this area or other IP related areas.  So thank you for being here.

Keyes

[34:54] I would really enjoy that Kent.  So I’ll look forward to next time and thanks again for having me.  Always a pleasure.

Schmidt

[35:00] As always I’m indebted to the extraordinary team at Dorsey for making this podcast and episode possible.  For more resources on this and other litigation risks, go to litigationrisks.com where more information can be found including a book on managing litigation risk written by yours truly.  Until next time my friends this is yet another reminder that are there are a lot of sharks swimming out there in the murky waters so swim safely.