Can You Copyright AI-Generated Work? Tech, IP and Entertainment Lawyer Lisa Oratz Explains

Orlando (00:00.519)
so wait, I should use senior counsel or not?

Lisa Oratz (00:03.917)
You can use it. That’s fine. That’s my title. I just say it makes me feel

Orlando (00:08.526)
Okay. I well, I promise you you won’t. We’re not going to do anything else to make you feel old. I promise. wonderful. Well, if you’re ready, I’ll get started. Awesome. Lisa Oratz, thank you so much for joining us on the podcast. We’ve been trying to do this for a while. I’ve known of you for quite some time. I’m really excited to have you. thanks for being on the podcast.

Lisa Oratz (00:14.269)
Thank you.

Lisa Oratz (00:18.673)
Yep, sounds great.

Lisa Oratz (00:32.263)
Well, thanks for having me. I’m very excited to be here.

Orlando (00:34.94)
well, great. Well, Lisa Oritz, you are senior counsel at Ash Hurst Ashurst Perkins CUI. and you first came on my radar when I saw you speak at AI on the lot, which is a great conference out in LA about technology in entertainment. and you know, you’ve done some amazing things about trying to organize and and and help people understand a lot of the copyright a copyright.

Lisa Oratz (01:00.541)
The copyrights to

Orlando (01:04.366)
Claims, situations, positioning around especially AI. so you know, we have viewers across film and television and advertising. You know, could you you know, for someone outside the legal world, what do they need to understand about what you do and who you advise and the problems that you’re helping clients solve?

Lisa Oratz (01:25.009)
So my work covers three areas, technology law, entertainment, and intellectual property. So my clients range from big tech companies to little startups, but I also spend a lot of time working with studios, production companies, and artists on the entertainment side. And my real special sort of fun spot is where these all collide. and some some of the issues that we were talking about AI on the lot is really at that intersection. So I advise a lot of clients on how to address some.

of the you know issues and navigate this uncertain legal landscape on on IP issues. But I also work on deals, contracts, you know really a wide range of work for you know entertainment, advertising, gaming, and tech clients.

Orlando (02:12.193)
Yeah. You know, I I think that, you know, there’s there’s a lot of conversations around copyright challenges with AI. What do you see most people get wrong when they you know when they only have a surface level understanding of the copyright challenges that we’re facing here?

Lisa Oratz (02:29.457)
Yeah, and there is a I I’m glad you’re asked because there are lot of misunderstandings and misinformation out there. You know, it’s a very complicated area and everybody sort of jumps in and wants to be an expert. but there’s two key things that I really want, and we can dive down in more detail as we talk, but really want people to come away with is, you know, there’s the copyrightability and protectability aspect. And there were a lot of articles and fear out there in in with respect to some of the recent decisions came out that says, you know, AI generate

Work isn’t protectable. But that really is a misunderstanding. There’s a lot of question about exactly what’s protectable, and in particular, whether the actual output that comes out of an AI tool is protectable. But the law is still developing on that. And even with that uncertainty, and frankly, a copyright office that is a little bit hostile to AI. You know, the works, you know, the general works as a whole, so if you’re whatever you’re creating, as long as there is sufficient human authorship, which is the big question.

Orlando (03:27.991)
Mm. Yeah.

Lisa Oratz (03:29.441)
These things can be protectable. So that’s one misunderstanding. And the other one is around infringement. I mean there’s over a hundred cases that have been brought now. But from a user standpoint in particular, there’s a lot of things that users can do to really avoid and minimize any risks of infringement. And again, it’s an area where with a lot of uncertainty, but where you’ll see a lot of, you know, sort of alarming headlines, particularly with some of the we had a there was a recent settlement with Anthropic, which we can talk about.

But there’s a lot of of you know misunderstanding exactly what the parameters of that decision mean. So when I talk at these conferences and podcasts and things like this, my goal is to really help people understand the line between what we know about the law, what we don’t know, and how you can navigate that in an environment that frankly is gonna be uncertain for quite some time, likely.

Orlando (04:21.889)
Yeah. And I think that, you know, you you referred to it there briefly, but I you know, I think that there’s obviously the fundamental copyright provenance issues with the models themselves, their training data, and what’s called fair use, the fair use argument that similar to if I were to read a book and then be inspired by that book and then you know, create a book myself that’s that’s inspired by that. I that’s a fair use of that.

Copyright. And then the secondary point that you’re making as well, which is then what happens when in a human workflow, that stuff that comes out, you mentioned the hostile copyright office. You know, what does one need to do to actually have ownership and have an IP protection or copyright protection around that outcome of that?

Lisa Oratz (05:17.287)
Yes. So maybe talking taking those sort of in in reverse order, perhaps. so one of the key issues and probably the most concern that I hear people is if I use AI, am I gonna lose protection? Am I gonna be able to protect and prevent other people from using my work? Excuse me. And the big issue here is that there does have to be some human authorship in order to have a copyright. And we’re still trying to figure out, you know, exactly what is sufficient if you have very detailed prompt.

Is that enough? If you, you know, and and that’s sort of the big question mark where the copyright office has said that a text prompt by itself is not sufficient. It’s important to note that what the Copyright Office says is not law. We don’t know until we have actual case law. But what they say is important. I mean, they get to control what gets registered and they can be influential. But we don’t know whether based on a simple text prompt or even a complex text prompt, the actual content that’s

Spits out of the tool will be protected. But most of my clients, at least, don’t just put in one prompt and outspits an ad or a film or a creative work. They are doing a lot more with that. They are taking that, they’re combining different output, they’re modifying it. And even with a pretty narrow view of protectability under the Copyright Office view, they have recognized that these works as a whole are protectable. And for example, you know, we have worked on the film.

Critters and obtained copyright protection for that film, which was an AI-generated short film. It’s now being made into a full-length feature. But we we think this was the first actual copyright registration for an animated short film. And we were able to get protection on the film as a whole. And you do have to describe when you file a copyright.

what your what the human authorship of that can consists of. You don’t have to go into great detail, so you don’t have to give every element of human involvement. But you do want to make sure that when you create works that you document, that you insert a lot of human involvement, you want to make sure that you’re taking work, you’re combining that. Because the Copyright Office recognizes, and copyright law I should say, recognizes the concept that when you even if you take

Lisa Oratz (07:46.869)
Unprotectable elements. When you make selections and arrangement of those elements and put it into an end product, that end product is protectable. So on that basis, it’s not hard to get copyright protection for you know that output. And we’ve done this now several times. So that’s what I really want people to come away with. So you can protect the output. And what we suggest in terms of trying to address the issue: well, what about the question of the output that comes directly out of machine?

Because theoretically, the concern is if somebody were to extract that output and claim it’s not protectable, could they use it? But you don’t have to tell them what was the original output and what did you change and modify. So if you’re taking output, you’re combining it, you’re changing it, you’re putting it into your film. It’s very, you know, it’s very difficult as a practical matter for somebody to figure out what might be unprotectable elements and be able to take them and use them. So I think a lot of concerned.

concerns as a practical matter really aren’t a big deal.

Orlando (08:50.059)
Yeah, I think that that’s what I’m I’m intrigued by your work is that most of the public conversation around AI and copyright is framed as AI work can’t be copyrighted or it’s tainted somehow. But your work seems to sit in the more important middle when AI is part of a human creative process. What evidence proves that the human contribution still matters? and I think that that’s that’s the more adult conversation I think to be having. And I’m so glad you brought up critters because

I don’t know if the you know our audience will remember, but critters was one of the first, certainly the first online piece of work that I had seen that now everybody would call slop. but what that what I take that to mean is that it was end-to-end AI creation. it was these cute little critters in a forest. and then it was the first thing to be optioned as an AI piece of work, and you you

Created the copyright protections around it. So I guess the question, and just using critters as an example, what was the human creative contribution or the meaningful human intervention that differentiated critters and made it both not slop, but also a professional project that could then be protected?

Lisa Oratz (10:09.905)
Yeah, because I I definitely wanted to make sure we disagree with the the characterization this ass off because I

Orlando (10:13.845)
Yes. Mm-hmm.

Lisa Oratz (10:14.545)
very high quality output. But the thing that was important is there was a lot of human input in that. So and what’s interesting about the short is it was in there, you know, we didn’t have SORA, we didn’t really have when I should say we’ve there there weren’t a lot AI video tools that were broadly available. So that was actually created using Dolly and using still images. So the images, the character images, the background images, those were all created through DOLI. But once they had those images

They combined them. So for example, there were a lot background panels. They were put together, sort of stitched together, modified, sprinkled with pixie dust. I love that thing because literally you could see that on there, and that was something we used to describe them. So there was a human that that went through and did all of that work. And then the actual character images, those were all modified using, you know, using tools that a human was involved with.

human script, human actors. They captured, you know, so the they did sort of a voice capture. So there was an actual lot of human involvement in that, which is one of the things that made it fairly easy to obtain protection.

Orlando (11:28.321)
Hmm. well that’s a ma amazing. I’ve I I haven’t heard it heard it characterized as as fairly easy. That’s great. what what had to be documented then on their side in order for that to to hold water?

Lisa Oratz (11:43.805)
Mm-hmm.

So I mean it’s always good to internally document all that, but when you actually apply, a lot of people think you have to give a great amount of detail to the Copyright Office is exactly what all that involvement was. And they really the Copyright Office doesn’t want a lot of detail. They want to know that. So we described it really just a paragraph or two of and sort of exactly what I just said was we explained how the images were generated, they were modified, they were select this selected and arranged in a particular manner. There was a

human script. so we were able to describe that. and we always recommend again keeping as detailed records as you can in terms of all those human touch points, just because there’s so much because we don’t actually

Orlando (12:26.123)
Is that logging every prompt and output and then logging, you know, I mean, is it is it a huge logging task? Is that that what you mean?

Lisa Oratz (12:33.747)
I mean i it can be, you know, because you know, i for many reasons it’s good to log prompts and to keep a record of that. but as you remember the copyright office, as I said, doesn’t really recognize the copyrightable the output based on prompts. So the thing that’s the most important is what you do with the output, how you modify it, how you combine it, and all of those big but I do think that all of it should be documented. We don’t y know exactly what’s going to be the most important.

And one thing we did which was kind of interesting and in critters and we’ve done this in subsequent applications as well is you know you have to disclaim what’s not protectable. The challenge is we don’t know legally what’s not protectable. But we know that if we didn’t mention the actual output, if we tried to claim that, the copyright office would have rejected our application. But at the same time, we didn’t want to exclude the AI-generated images from that protection because again, they could be wrong.

And we don’t know until we have some case law. So what we did is we excluded AI-generated images to the extent not copyrightable. So it’s kind of a circular argument to give it a little bit of room to flex as the law developed. We didn’t really know what they did with it, but they accepted that. So, you know, in this case at least, it’s not, you know, there there’s gonna be uncertainty about those initial images that came out. So those are the initial character images, the initial background panels.

But to the extent that we actually and we should have a decision coming up sometime soon because there was a recent copyright office ruling that had denied protection based on text prompts that was challenged. And so we may actually get a little bit of law on this. but until that happens, you know, we have at least a claim that, you know, even those text prompts are protectable. But the most important thing that I really want people to come away with is that as long as you make sure that you do include

you know, humans, you know, they talk about humans in the loop, but they’re making decisions, they’re shaping their because the what the copyright office view is that they want to make sure there’s a human sort of controlling the output. So you want to show that there’s a human being sort of in charge. whether or not we that it has to reach that higher threshold, we don’t know. But if it does, you know, that you’ve showed that. So we always suggest try to make sure you reach that bar. hopefully we’ll get some clarity that that’s not needed. But the document

Orlando (15:00.109)
Mm-hmm.

Lisa Oratz (15:03.327)
Documentation, again, it’s important, but you don’t want to share that, you don’t want to make it publicly available. Because you don’t want people to know what you cr what came out of an AI tool and may or may not be protectable and what you add.

Orlando (15:17.441)
Yeah, it’s interesting because we we have a lot of people who who watch this who are creative professionals. So I think many of them will be encouraged by what you just said because they’re used to working in very human-centric ways and and producing in in you know the the and and producing high-level professional work that has lots of human intervention. And so I think that they’re used to that. I think they’ll be very comforted by that. I think that

It’s the kind of creators right now who need to hear a lot more of this because professional creators, you know, and professionals within film industry, advertising, et cetera, they’re used to having a business affairs department that you’re running these sorts of questions by. whereas it’s actually the the new creators who are, for the first time ever in their lives, having access to really to highly developed post-production tools, they need to hear a lot of this as well to to

help them understand the rigor that goes into creating something that’s professional and defensible.

Lisa Oratz (16:18.107)
Yeah, and the other thing I want to mention that helps in terms of again maximizing your arguments for protection is start with stuff other than text prompts. And again, text prompts may be fine, but a lot of people start with sketches, storyboards, and to the extent that you start with something like that that was clear that those are clearly copyrighted. If you create a sketch, you create a storyboard, that’s a copyrightable document in and of itself. So if you’re starting with that, to the extent that that’s reflected in your

output that is going to be protectable. So that’s an additional thing that people can do to make sure that that they’re maximizing exactly what’s protectable is to start not just with text prompts but with creative what they call expressive prompts. Very interesting

Orlando (17:05.005)
And so that’s very interesting because where do we then draw that line? obviously, if I’ve drawn something or I’m a storyboard artist, great. But many production designers begin with a mood board. So if I bring to the LLM a mood board that that clearly defines an artistic area I want to explore from a design sense, and then I use that as a jumping off point for for my work, is that copyrightable?

Lisa Oratz (17:35.033)
And you know, and part of it depending exactly what that design board means. The trick about copyright is you cannot protect ideas. You c protect ex the way those ideas are expressed. So to the extent that this is at a very high level and it’s merely ideas. And the same thing would be true with very simple text prompts, even you know, under a more lenient standard of what’s protectable. If you just have a very simple prompt or a very simple design idea, that’s not protectable.

The di tricky part is where do you go, where’s the line between what’s an idea and what’s the specific expression of that idea. So something like a drawing tends to be pretty clear where you have a design board, you know, it might depend specifically on what type of things are in there. But I still think it’s helpful to the extent that you can show that there was this human input that’s shaping and determining what the ultimate output looks like.

Orlando (18:31.499)
Mm-hmm. That’s very interesting. Yeah, because you know, w where do you think

And I know I’m sure this is very hard to to pinpoint. Is there a threshold that you think

defines that at all? You know, is is there a yeah, is there a threshold that is a good rule of thumb, let’s say.

Lisa Oratz (18:59.033)
that is very tricky, you know, ’cause again it depends, you know, like you can think for example, you think about just a movie in general. You know, there are general themes of movies, you know, boy meets girl, meet cute, and everything that these that are merely ideas. But as you start to get more specific about plot elements, how they meet, what they say, what they do. So the at the one end, you know, you have a script with very specific words and expression and direction. At the other end you have these broad ideas. It is very hard to say where you’ve crossed that line.

line. but definitely would need to be something very specific to what you’re doing and something more than just the mere idea of, you know, again a you know, so you know, p boy goes into a cafe, sits down, that’s pretty high. But if you, you know, get into it certainly dialogue, you’re getting into

you know, drawings that show w how they do it, what they look like, that’s sort of the expression, the text, the images, and a lot of that detail. But again I want to remind you that it’s important that even for unprotectable elements, the way that you stitch that together, the way you combine those elements is itself protectable.

Orlando (20:11.949)
Right. Yeah, because I think that that’s that’s a that’s very interesting. Yeah, well, because I was trying to think, you know, at a certain point, is there a sort of level of developed prompt where that prompt does become protectable? I mean, at a certain point, if I’m writing a prompt, there’s a point at which that prompt becomes an essay, and that essay becomes a thesis paper, and that thesis paper becomes a novel. You know, so you know.

Lisa Oratz (20:38.055)
Yeah. And that’s a function.

And that’s what’s interesting about the Copyright Office view on this so far, because they said even if the prompts themselves are detailed enough to be protectable, somehow they’re viewing those prompts as direction, as if for example, you’re if you hire somebody and you give them instructions as opposed to the content. Now I would make a different argument because I think i again a very simple prompt, save an image, you know, give me an image of a teddy bear sitting on a blanket, that’s not protectable. But as you as

Mentioned, if you start to get very detailed and you start to describe in very granular detail exactly what that should look like, I think there are arguments that that should be sufficient human authorship. But in some of the initial rulings that the Copyright Office made, some of you may be familiar with the Zarya of the Dawn graphic novel. This was the first decision that the Copyright Office made. And they did show that they were very, very, you know, it wasn’t just, you know, you put in a prompt and out came a picture. They regenerated. and these prompts, some

Some of them were a hundred words or more. And they, you know, they made the argument. I thought it was a pretty convincing argument that it was detailed enough to really shape and control the output. Because that’s what the copyright office seems to want, is that it controls the output. Although in other areas we don’t really see that, and other rulings outside of AI, which is why I say I think they’re a little hostile to AI. But you know, where you get detailed enough, I think you should have arguments. And I mentioned there was this another

ruling which involved an image that actually won a prize at a at a the Colorado State Fair. They also got this ruling that, you know, even though they generated and regenerated and and and had very detailed prompts, that wasn’t protectable. And they’ve challenged this now in a court case. So I’m hoping that, you know, hopefully by the end of the year, because it’s all been fully briefed, some of the initial arguments, that we’ll have a decision. And I think it I would not be surprised to see a court say that

Lisa Oratz (22:40.397)
you know, at least a very, very detailed prompt. ‘C you could have a prompt that was so detailed that it is really controlling every aspect of what that output looks like. But we’ll have to see what you know what the court says ’cause the Copyright Office so far has just, you know, had a a pretty clear direction that they don’t think prompts alone are sufficient. But again, that’s law, it’s just their opinion.

Orlando (22:49.143)
Mm. Yeah.

Orlando (23:01.751)
Right.

Yeah. You mentioned something just before where you said that editing together elements is is protectable. So the the human authorship of editing and choosing amongst many different things, does that mean I could have edited together a series of things, but I could not own the characters? Is that where, you know, so I can own the final video if I’ve edited together lots of things.

because that’s my human authorship, but I don’t actually own any of the specific elements within that edit.

Lisa Oratz (23:35.175)
Yeah. So if we assume that the individual elements, so let’s say a character that was AI generated, no modifications, and again this is the copyright office view, I don’t know that they’re right, but if we assume they’re right and a particular character or a particular element isn’t protectable, if you select which ones you want to use and how you arrange them, that work as a whole, so that selection and arrangement is protectable, but that doesn’t protect, as you noted, the individual unprotectable elements. You don’t suddenly make them protectable.

So if that’s the case and those particular characters weren’t protected, theoretically you could take those individually. And if you could figure out what isn’t protectable, you could take that and use it. And that’s why, you know, when we advise clients, we say, you know, if you generate something particularly through a text prompt. Well, first of all, use those expressive prompts, because then those characters will have your content in them and that there won’t be as much of a question. But let’s say you take a very simple text prompt and outcomes a very simple

character, take that, modify it, because the modified version will be protectable. Now the original version won’t be, but a user isn’t gonna know what was the original version looked like, what did you modify? So as a practical matter, it would be hard for them to figure that out. But if they could, yes, they could take out those unprotectable elements individually and use them, as long as they’re not copying the way that you selected and arranged them into that final.

Orlando (25:02.721)
Yeah, which in something like James Bond, for example, James Bond himself, if that weren’t protected, anybody could be making a James Bond film, all the character names and and likenesses and all those sorts of things. and so it’s all well and good to have owned you know, Casino Royale or or Goldfinger, whatever the first Bond film is. But if you don’t own all the underlying characters enough, you’ve created one great thing, but everything else is sort of a free-for-all.

Lisa Oratz (25:31.987)
Yeah, and there are, you know, there are public domain characters that people write a work around and they don’t have the exclusive right to those characters, but they do have, you know, when they put a script together, when they put you know, when they film and they have very specific elements in traditional media that would be protectable. So it’s somewhat analogous. You know, here you have the additional layer that, you know, when you you know when you’re filming, when you’re generating those images, there’s this question mark about to what extent. but

That’s why we encourage people to make as many modifications as possible and to insert in your prompting process, first of all, even though we don’t know if it’s protectable, make those prompts as detailed as possible. Make the argument that you are, you know, truly shaping what that final product is. is as much as possible, add those sort of expressive prompts in addition. and I would say most of my clients do that.

Orlando (26:29.229)
Right. I mean, I’d love to get into you. You know, you’ve mentioned some of the decisions that have been coming down the pike about all of this. I mean, obviously, when I first saw you saw you talk, there hadn’t been any decisions on fair use in general. the anthropic case hadn’t occurred. It seems to me when all these things happen, on the one hand, you get answers and guidance for this is the way everybody sees things now. And then very often it just raises a whole bunch more questions.

Lisa Oratz (26:59.078)
Absolutely.

Orlando (26:59.361)
Where are we right now in terms of the decisions that have been made on both well, what where are we right now? Can you give us a sort of update on where we are regulatorily regarding, you know, AI, copyright, and fair use?

Lisa Oratz (27:06.566)
Okay.

Lisa Oratz (27:13.971)
Yeah, because this is, you one question is do I only the other question is if I use this, am I going to infringe somebody’s right? and you know, I’ve spoken on this topic and AI and copyright for, you know, over a dozen years, and for a long time I I had to say, well, there are no cases. And we’ve kind of looked at some analogous cases, analogous cases. But we do have a couple of cases now, or you know, three to be exact, and two that specifically deal with generative AI. so we have a little bit of information, but as you know, it sometimes it just raises more questions.

so these two c and these two cases came out within three days of each other last year, and that’s the Bartz versus anthropic case and the cadre versus meta case. And just to set the stage on what the question is, is you mentioned fair use, and for people who may not be familiar with the concept, and there’s a lot of sort of misunderstanding about fair use. But fair use is a concept that certain types of uses are permissible for things like criticism, comment, news reporting, etc. such that they

That use is not copyright infringement. And it’s a balancing test, which is very difficult because it’s very fact-specific, but there’s, and I won’t get into a lot of details, but there’s four different factors you look at. And there’s two particularly important factors that I will mention. The first one is the nature and and purpose of the work and whether it is what they call transformative. and that means does it really just, you know, f function for the sort of the same purpose as the underlying work, or does it add something new? Does it make

you know, create something that is of a different nature or purpose or character. The other factor that’s important in particular is the impact of this use on the market for the work. So does it really harm the market for the work? And so, you know, when you do a fair use analysis, you’re looking at, you know, though those and the other two factors, which are the amount, the amount that you use and kind of the nature of the work. So in the Barthes case, which was the first one to come out, well actually

In both these cases, they found fair use on the specific facts. But it’s a very it’s that doesn’t mean that the that this issue is you know resolved because there were a lot of limitations and caveats in these cases that that give us a lot of pause. but one of the important things that in the similarities in these cases is that both cases found that AI use is very transformative, which is important because that’s one of the big factors. And in fact, one of the courts said it’s probably the most transformative.

Lisa Oratz (29:43.673)
thing in our lifetime. And the idea is that you’re using this because you’re not, is that you’re making copies and the purpose of those copies is, you know, is not to make new copies, but to train, you know, when we’re talking training. And I mentioned both these cases are focused on the training side, which is the copies that you make to create the tools, not output, which is a different issue when you output issues. But in these cases I said this was highly transformative and both found fair use on the

Orlando (29:57.837)
Mm-hmm.

Lisa Oratz (30:13.417)
Facts. but in the Bart’s case, the distinction they made, they said, well, training use is okay, but they had two issues with what Anthropic had done. One is that they said that when they created this training database, they copied from pirated sites. So they didn’t have know not just not unlicensed, but these were pirated sites, and that they also used this not just for training, but they created a central library of works that they used.

used for training and other purposes. So because of that they found that those particular uses were not fair use and that’s where we got up

Orlando (30:45.133)
Mm-hmm.

Orlando (30:51.735)
So the issue there was that they hadn’t paid in the first instance, they hadn’t actually paid for the work itself. And so that was considered theft because, you know, just like when we used to rent movies at Blockbuster, there was that FBI thing, you know, you you can’t share this, you can’t exit publicly exhibit it. you know, the the you know, the the

Lisa Oratz (31:13.687)
But not it’s but not qu yeah, not quite. And this is where the nuance comes in, because they said it was pirated, which is not the same as unlicensed. because for example, if you bought a so you know, if you paid if you bought a book and you use that to train, that’s not pirated. and even though you didn’t, you know, it’s important to know that doesn’t mean you were licensed to use it for training, but it wasn’t pirated. But there could also be other qu and there’s a question about what pirated means. Let’s say you access it

Orlando (31:16.619)
Right. Sure. Right. Right.

Lisa Oratz (31:43.523)
lawfully but you didn’t pay for it. So there’s still a lot of questions about what you know what pirated limits. But at least if you go because they went to these really well-known sort of pirate websites that have pirated content. But the other piece of it was the fact that they used it not just for training but for other purposes. And there’s some question about well if it was was used just for training would that have been okay? But the case did settle for you know 1.5 billion and that’s where you got a lot

of things that said, now if you don’t license content for training, you’re gonna, but again it didn’t say you have to license it, it just said really it should be non-pirated and used only for training. but the interesting thing is so there, you know, at that judge that decision, but in the second case, the cadre case, the judge commented that he didn’t care if it was pirated, that this was fair use, it didn’t matter. So they disagreed on that point. At least, you know, one judge has said that it does matter how and where you got the content.

Orlando (32:38.381)
Mm-hmm.

Lisa Oratz (32:43.337)
At least if you use it for things other than training. But in the cadre decision, which was the second decision, again, he didn’t care about the pirated issue, but his concern was in that last factor, which was the market harm. And he recognized a theory that sort of it’s called indirect market substitution, which is the fact that if you flood the market with AI generated works, that might harm the the market even if the works individually aren’t infringing. Now in this case he said he didn’t.

didn’t find that because they didn’t prove their case. But he suggested that others could and kind of gave a little bit of a roadmap of sorts of how others might make that claim.

Orlando (33:23.413)
And this is the principle that, you know, it’s it’s fine for me to take a a book, be inspired by it, and write it another book, but when an AI does it, it can potenti potentially be writing a hundred other books that are similar to that book at the same time. And that competes with me as a writer who I I have a finite ability to output books on my own as a writer, but AI has a theoretically infinite ability to compete against that writer. Is that right?

Lisa Oratz (33:50.821)
Right. So there’s this notion that it’s different because, you know, there’s you know

toolcomes will make the argument that it’s the analogy is the same and so this argument well no it’s different just because of the scale. but just like the cadre judge disagreed with the Barts judge, the Barts judge disagreed with the Cadre case, because this indirect market substitution theory had been raised by the copyright office and said, no, that that doesn’t matter. That’s just competition. So again we have two different cases that came out within a few days of each other, both of which had very, very different reasoning. So I

It’s gonna take a little more, and there’s you know a hundred more cases to follow. So it’s gonna take a little bit more to shake out exactly what those parameters doesn’t matter if the prop content is pirated, does it matter, you know, if market substitution you know is indirect and and that the fact that you know you just the mass scale of AI. And so bottom line is we don’t know. But one thing I also wanted to mention is that this involves the training piece. And that really doesn’t directly impact users.

Because the training issue has to do with how they train it. There are a number of cases that were coming down the pipe around output, but we haven’t had decisions. And that is going to really be a matter of whether your output is substantially similar to some other work, likely something that was in that training data.

Orlando (35:16.013)
So this is the these are the cases like Warner Brothers Disney and mid have sued Mid Journey along with with NBC Universal because they recognize that if you had at that point in time had prompted Mid Journey to show you Darth Vader, it would show you a a character substantially similar to Darth Vader that was clearly referencing their owned character. Is that right?

Lisa Oratz (35:41.199)
Yes, but a lot of just to to clarify a lot of those cases are still about the training and they’re showing examples of how the training how the output shows up you know, the prompt to show that the train you know, that the training is in fair use. Some of the cases, not all of them, also address output. but a lot of these a lot of the cases are focused really solely on the training piece and all of the cases so far have only been against tool providers. We don’t yet have a case that’s really

been brought against, you know, a studio or somebody that or that’s using AI as a tool that’s not training models, but that is using that as output. And that’s, you know, the other, you know, sort of practical advice that I’m often working with clients is as a user of an AI tool, you don’t generally, and you know, all of this is still

Orlando (36:33.643)
You’re okay. Yeah.

Lisa Oratz (36:34.587)
Out in the law, but you know, you’re not gonna generally it’s unlikely that you’re gonna be responsible for the training. What you need to worry about is making sure your output isn’t similar and you have a lot of control over that because you know, a lot of these cases where they gave examples, those are highly orchestrated examples intended to generate similar content. you are prompting things and trying to emulate something, you know, give me this, you know, in the style of you know, this artist, even though style itself isn’t actually protective.

If you reference IP in your prompts, you have a greater risk that you will somehow, even inadvertently, end up with output that might be too close to another work. So if you avoid doing that, well, first of all, if you start with those expressive prompts, drawings and things that you create, as a practical and as a technical matter, that’s likely to dominate the output in a way that it’s unlikely that inadvertently some you know random word.

work and the training data is going to show up in your output. And even if you, you know, and if you prompt things in a way that’s generic and you’re not trying to emulate and get output that looks like someone else’s, I just think that the practical risk that you’re going to out of all the images that we’re trained on, that you’re going to pull up something that happens to be substantially similar to a work that was in the training data, you know, is not as likely. And you know, I’ll also mention that there are things that

Orlando (38:00.632)
Yeah.

Lisa Oratz (38:04.287)
You can do as a user too to try to limit that risk when you choose tools that you use. You know, there are tools, you know, most of these tools, at least in an enterprise version, also have indemnities around use. Now, those indemnities have limits and caveats and restrictions, and so you need to look at that and make sure that you, you know, are kind of following that. and that can be something when you’re choosing tools to look at. But I think, you know, users, while we can’t say that them, you know, in an uncertain legal

environment that the risk is zero. I think that the practical risk, if you’re not trying to have the output look like something else and you take reasonable measures to avoid that, the risk that you’re going to end up with something that somehow infringes inadvertently is pretty low.

Orlando (38:50.539)
Yeah, so I mean, you know, so so each of these decisions has sort of moved the conversations forward, brought up some some other questions. But can we fairly confidently say that the fair use argument has generally won? I mean, not without saying that’s not okay, that’s not okay, but the the central fair use argument that training on you know copyright data is sort of okay. I mean, as has that been

Lisa Oratz (39:19.611)
I think it’s true. I think it’s

Orlando (39:20.213)
Has that part of the argument been generally resolved?

Lisa Oratz (39:23.217)
I mean, I don’t yeah, it’s too early to say that because you know what we we know, we have two cases, first of all, only two cases, and you can have different circuits that disagree, you can have other cases that have totally different so you know it’s far too early to draw conclusions. and because as I mentioned, it’s such a fact specific analysis that the facts of each case can also mean that in one situation something’s fair use and not. is not. Now there have been some beneficial in these first two rulings, some beneficial

Things, one of which is this transformative factor, because that is very important. And the fact that both of these cases said it’s highly transformative is very helpful. The other thing that was very helpful is that in looking at that last factor, which is the impact of the market, both of these cases said that what you don’t look at is the market for licensing the work itself. Because there’s an argument that people make, and it’s kind of I’ve always thought it was very circular that, you know, if you say

you know if I had licensed this these materials to you for training you know I’ve now lost that income and so that’s market harm but that’s exactly the light the the harm you’re trying to figure out if you need to license so it’s always a bit circular to count that and both of these cases did say no you don’t look at the you know at the loss of marketing from licensing the thing we’re trying to decide if you have to license so I think these cases both

Had some beneficial rulings. But because you know, i in in the Bart’s case you had this question of you know you still have to look at where the content comes from. In Cowperay, you still had this question of, you know, indirect market substitution. from that, you know, there were enough caveats in that that until we have a little more case law coming out of this, you know, we’re really not gonna know. And I frankly think it’s gonna be, you know, quite a while because there’s so many cases, each of which could could

come out differently. You can have different circuits, different parts of the country, you know, might rule differently. And these are lower court decisions too. So again, we we’re we’re we’re not very close to re resolving the core issue, which is is you know, is training on copyrighted material fair use? You know, we don’t really have a you know a clear answer

Orlando (41:41.91)
Mm-hmm.

Okay, so a few early wins, but the game isn’t over.

Lisa Oratz (41:48.647)
Yeah, and I’d say a few early wins, but a few early caveatted limited wins. Not uncle knows unclear victory for either side. So there’s something for everyone to not love.

Orlando (41:53.537)
Right, okay.

Right. Well, I’m just just because they didn’t fully win, I’m sure that won’t stop the the tech companies from declaring victory. I’m sure yeah. all right. Well, you know, that’s very interesting. Does it feel a little bit like the creator, the end creator, if they’re an individual and not an enterprise or not a brand, is generally

Protect seems to be at the moment, whether it’s just because they’re too small for I or not, generally seems to be pretty protected in terms of their freedom to use post and these things, so long as they’re not monetizing something that is that is potentially infringing. But they sort of have a certain degree of freedom to to create at least.

Lisa Oratz (42:45.073)
I’m not quite sure if I’m if I’m following your argument, but you know, the fact that if one of the factors is, you know, if it’s a kind of a non profit use versus a for profit, but that’s really not a key factor. I think for you people who are using creatives who are creating materials, there is you know, if if you are outputting as part of something that is substantially similar to an existing work, there is risk. But my you know, but my advice to them

is to take steps to make sure that you’re not doing that.

Orlando (43:19.245)
Yeah, because I I think that one of the things that I keep coming back to is there’s a there’s an online creator who’s done something called Star Wars Beggar’s Canyon. And it’s an AI piece, that’s a a Luke Skywalker story, I think, in between, you know, some of the films. And he’s sort of decided to take it upon himself. I’ve been surprised at how at you know, with so many big companies taking things down, I’ve been shocked that Disney hasn’t sent a cease and desist or hasn’t

You know, pressured YouTube to stop it. and I think it’s because people really quite like it. I think it’s, you know, at a time where, you know, the the Star Wars franchise is losing a lot of brand relevance, this is gaining quite a lot. And I think that there’s a degree to which, so long as people aren’t, you know, he’s not going to a movie theater and exhibiting it around his town, but he’s just putting it on YouTube. I think that there’s a degree to which.

Lisa Oratz (44:03.143)
Mm-hmm.

Orlando (44:19.009)
You know, certain creators are are being allowed to just go, but you’re like you said, it’s not without risk. He is still at risk of them sending a cease and desist and taking that hard work that he’s done down.

Lisa Oratz (44:23.313)
Okay, so that’s the yeah, that’s management.

Lisa Oratz (44:29.905)
Yeah. And and that helps me frame a little bit more what you’re asking about. So fan fiction’s always an interesting area for companies in terms of policing. And you know many people know Disney tends to be very aggressive in policing things. So if it is interesting,

their view here. But with fanfiction, you know, a company who’s a rights holder has to determine, you know, whether this is, you know, something that’s beneficial or not. And and this is where when you get into this again, fair use, you know, all these factors, if it is not, you know, if it’s not monetized, if it’s not for profit, that is one part of the fair use analysis. But you know, things that aren’t for profit, if you’re still, you know, if you’re still copying material taking content, you know, there’s still

Orlando (44:51.255)
Mm-hmm.

Lisa Oratz (45:12.157)
And I say arguments because again you have depends on how much you use, how you use it, so many factors. But I think ultimately the decision of a company of whether to go after people who are doing fan fiction is a practical risk. So it’s not that it isn’t infringing necessarily. It may be that they think it benefits the brand so that they tolerate it to some degree. but it’s a it’s a tricky for for brands and and IP not and IP owners

it’s a delicate balance because if you don’t, you know, if you want to make sure if you enforce it, is that gonna affect affect ability to enforce for other types of uses you wanna make. But fan fiction’s always been, you know, even outside of, you know, anything with AI has always been an area where content owners are very careful about what they do because they do wanna they can be very beneficial to have this, but it can cross a line too. So

Orlando (46:07.915)
Yeah. Well, it’s funny because before AI came out, it felt like every big brand and media company was obsessed with figuring out user-generated content. They were like, great, we want to give them assets. We want them to post online. And then suddenly AI allowed them to do it at scale. And suddenly, you know, suddenly the big media companies became very nervous about U UGC. And, you know, and then the decision was sort of made for them because all of their stuff was in these models. so it’s it’s been interesting to see them wrestle with that.

Lisa Oratz (46:35.345)
Yeah. Well here’s the other interesting thing is, you know, you have some people who are really and

really in the middle on both sides. So you may be a content owner that’s very interested in protecting your stuff. At the same time, you know, particularly like studios, you may be very interested in using AI and and all the efficiencies and other things that it can bring. So you know kind of you you’re in this middle because you’re not just a content owner, you also content creator that wants to use it. And so it’s been interesting to see, you know, the views of the studios change. And I think, you know, although we still have all these cases on training

A lot of the conversations I’m hearing are really focusing on that output question because there may be, you know, and I I like obviously can’t speak for the studios, I don’t re not speaking on behalf of, you know, anybody myself. But I think there are questions about you know, the training. Again, we still have these cases, but you know, the output and making sure that you’re protecting so, you know, you mentioned, you know, Disney that that people aren’t generating stuff that looks like their content is

a far more critical issue than the training issue. I’m not saying they’re giving up on the training issues, but I think I’ve just hearing conversations really focusing a lot on output and less on training than I did say a year ago.

Orlando (47:59.774)
fa that’s fascinating. the the I I wanted to kind of, you know, we I I talked very briefly there about the end creator. And I guess what I’m want to try to unpack with you is obviously there’s a lot of fear around professional creators, feeling like there are so many people coming into it. There’s a lot of

You know, there’s a lot of concern around the fair use i argument and job displacement and and things like that. You know, what is there a role

I’m trying to think how to put the question.

Should people be thinking about this legally? Are there basic things that they can be doing to protect themselves and their work? even at the most base level, I’m I’m somebody who’s been a creator, or I’ve been a somebody who’s worked in a creative company. I now want to kind of do something on my own. Is it possible to protect this thing on my own? Or must I be engaging with somebody like you who’s a

you know, a a copyright expert and a lawyer to help shepherd me through these protections.

Lisa Oratz (49:16.403)
Okay. I’m packing a lot there. So

Orlando (49:19.23)
Sorry.

Lisa Oratz (49:22.419)
’cause I’m not I guess I want to make sure if you’re talking about protecting their work. Because well let me just let me start with the beginning of your of your comments because it touched on something we haven’t really talked about, which is the job display. And so it’s not because we’ve talked about copyright, but one of the other issues that I talk about a lot and and focus on are the sort of digital replicas and this concern about talent. because there’s a lot of concern among talent that AI is going to

Orlando (49:48.331)
Mm.

Lisa Oratz (49:52.365)
replace them and a lot of so I speak on panels sometimes with Duncan Crab Chair Ireland, the head of SAGAFTRA and obviously I I

Orlando (49:53.751)
Mm-hmm.

Orlando (49:59.118)
No. He’s gonna be a guest in just a couple weeks. Yeah.

Lisa Oratz (50:03.281)
Yeah, Duncan’s great. I love Duncan. and and there’s obviously a lot of legitimate concerns that creators, that excuse me, the talent has. But I’d also say that, you know, i technology has been pr re you know, replacing people for years. You know, CGI, you know, we used to hand animate things, right? we used to hand screw bolts on cars. Technology has always replaced jobs. And this isn’t going away. and

Orlando (50:28.449)
Yeah, nobody gets their start in the mail room anymore.

Lisa Oratz (50:30.993)
Yeah. And there’s benefits too, you know, yeah, even for creators. I mean you can also, you know, if you create a digital, you may be able to make money without having to go in and spend a day of work, you know, and I know the units work very hard to make sure that people get compensated and have consent. but I do think that there’s a lot of fear of talent. But I think, you know, AI can also do something that that’s really beneficial for creators, which is, you know, reduce barriers to entry, you know, elevate diverse voices, because now you can you know, you can create a film.

Orlando (50:40.333)
Mm-hmm.

Lisa Oratz (51:00.977)
You can create content without having to have as much money and have a studio behind you. So, you know, there are always trade-offs on this. And while I’m sensitive to job displacement, I also think that this is something we’re just gonna have to figure out because it’s not going away. And, you know, we have, you know, if you use somebody’s there’s there’s protections if you’re using somebody’s replica. There’s also synthetic talent and the unions are trying to address that through rules.

Orlando (51:15.021)
Mm-hmm.

Lisa Oratz (51:30.837)
But you know, I think this is just

Orlando (51:32.683)
Yeah, that’s very interesting. I mean, because how with synthetic talent to to go back to the meaningful human intervention, how do you you know, if if that’s all synthetic, even if it’s like sort of performance capture, is that performance capture considered enough meaningful human intervention or?

Lisa Oratz (51:40.925)
Right.

Lisa Oratz (51:48.989)
Well we’re telling you we’re outside of copyright land now, so we’re not worried about

Orlando (51:51.819)
Right, okay.

Lisa Oratz (51:52.755)
We’re just talking about, you know, if a studio wants to instead of c hiring a real person, create, you know, the t this is the Tilly Norwood, you know, that got a lot of attention, create an only synthetic human being to to be a performer because then they don’t you know, you know, they’re not using someone’s likeness, they don’t have right of publicity issues and theoretically they can do that without paying talent. Now what the union has done in their most recent contract, they’re beginning to put provisions that require, if you’re a union signatory at least,

Orlando (52:01.973)
Mm-hmm. Yeah.

Lisa Oratz (52:22.709)
to make it as expensive to use synthetic talent as to use you know human talent by contractually requiring companies to you know only use synthetic talent if there’s a good reason to do that and to have to make payments. So contractually

Orlando (52:27.042)
Mm.

Orlando (52:37.303)
But is there an engineer at the end of the Tilly Norwood avatar that that gets the payment?

Lisa Oratz (52:44.379)
No, it’s not there it there really isn’t because it’s not I mean, you know, there’s a company there are tools that can be used, but you know, you’re not when you’re talking about rights, you know, person rights, rights of publicity, rights to use somebody’s, you know, name, image, likeness, there’s no person there. So, you know, just like if you create if you think about it in a video game, you create characters. They’re not real people. You don’t pay anybody because you created characters. This is a real this is like creating a c a video character, except it’s you know, it’s replicated.

And replacing what you might. But if you think about that’s what animation in a sense does, right? So it’s not a lot of stuff. I guess my point is it’s not really that different than we’ve done for many years. It’s just suddenly, you know, the scale of this and the concern seems different, but I would argue that it’s really no different than you know the synthesizer, the CGI, you know, factories that have, you know, you know, modernization does replace that.

We hope, and there’s many who argue that, you know, even though it might reduce certain opportunities, it also creates volumes of opportunities and new opportunities. So, you know, there’s a lot of arguments for pros and cons and people have to decide their own comfort with sort of the quote morality and ethics around, you know, AI use both in respect to copyright but also in respect to things like digital replicas. but you know, I still say this isn’t all that different than stuff that’s been going on for years, except that it’s at

Orlando (54:16.459)
Yeah, is there a concern for new actors potentially? Is that, you know, you could sort of be competing against Meryl Streep for the rest of your life. You know, like after she moves on and her digital self is there still competing for roles, you’re constantly being judged against the greatest who’s ever done it.

Lisa Oratz (54:35.035)
Mm-hmm.

Lisa Oratz (54:38.707)
I hadn’t thought about it that way, but you know, I mean I guess you know

I guess that’s true to some extent. You know, and this isn’t you know, obviously this is true across the board, not just in in entertainment, but you know, even lawyers, you know, we’re worried about lawyers replacing, you know, lawyers with all this. It’s just something as a society that you know, that we have to deal with. And but I do think that, you know, there is that it’s not g this is my personal view, it’s not going to replace real people. People there is a certain warmth and depth to a performance that at least at this point I don’t think you see from a lot of these AI

generated performances and I think people will still want to connect with specific actors so I don’t think it’s going to entire you we’re I don’t think in ten years or even in three you three years ten years we’re gonna suddenly have like no real actors and all talent’s gonna be AI but you know there definitely will be a role for that and help reduce costs and make people be able to make you know films and do things in films that otherwise they couldn’t do you know stunts there’s a lot of things that can be done

Orlando (55:32.758)
Mm.

Lisa Oratz (55:44.051)
more safely or scenes that would have just been ridiculously expensive to produce that now you can do. So you know, there’s always trade offs when you when you’re dealing with new technology.

Orlando (55:56.77)
Yeah, it’s it’s interesting. I mean, I I kind of yeah, it’s it’s so interesting kind of seeing everything that’s kind of happening now. I mean, my my thesis has always been that the the movies that we see, the 15 movies that Warner Brothers releases every year, the 15 that Paramount releases anymore every year, they’re still gonna release those, but it is gonna create a s a middle class of film that, you know, there’s there’s

Millions upon millions. I mean, only 1% of all the screenplays that have ever been written have been made into movies. You know, you’ve got an incredibly rich history of human creativity that’s just sitting there, that has never been activated, never been used, never been made into a film. And a lot of it’s probably very good. I’ve read spec scripts that are fantastic that I I’m fairly sure will never get made. AI sort of has the ability to take a lot of that and make it active again. And and

I’m I’m quite encouraged by that. I think that that you’re right. There will, you know, we still want to see people do things and human beings be expressive, and we still want to connect with that humanity. But in terms of stories being told, I think there’s a you know a whole level of films that will be end-to-end AI, films that’ll be hybrid productions, like the the one that The House of David, one that starred Ben Kingsley.

That was done on a volume, lots of AI in the back of that. Still done for about you know between five and ten million dollars in two weeks, human style production, but all on a stage. And then obviously the big budget productions. I just think it’s gonna, it’s just gonna broaden the the aperture.

Lisa Oratz (57:37.177)
Yeah.

And we’ve seen that, you know, given even putting AI aside, like as we started to get into the internet. I mean, you know, it used to be, you know, if you wanted entertainment, you had T V, you had movies. Well now look at this, we have videos and you know, young people today they don’t not that they don’t go to movies, but they spend a lot more time watching digital content and short form content, which did not exist, you know, fifteen years ago or however many years ago. So that was already a shift that and that has impacted t you know, film and television and all of this. But can we say that it’s you know, we don’t necessarily

Call that evil, but you know, AI, there’s definitely real people who who don’t like that. But this is just the evolution. You know, content has evolved. People want to maybe have more interaction with content. And I think this is yet another evolution of how we interact and the fact that, you know, and the same thing in the music industry. You know, it used to be you couldn’t put out a music unless you had a record company. Well now, you know, you can do a lot of things without needing to have a record contract. You can, you know, become a big you know, become a big star without ever having a contract if you, you know.

Orlando (58:30.881)
Yeah.

Lisa Oratz (58:37.777)
and put your content out on your own. So these are all shifts in this industry, you know, major shifts. but that’s you know that’s the same old story that that we’ve been seeing over time. Technology and changes, you know, are, you know, it can be monumental and, you know, it’s going to be hard to see, you know, what is the entertainment landscape going to look like in five years. You know, I think it’s hard to predict, but I will predict that we won’t see like no humans in content. Well there I don’t think we’re or all AI generation

Orlando (58:50.061)
Mm-hmm.

Orlando (59:04.299)
Yeah. Yeah.

Lisa Oratz (59:07.637)
you know, I think there will still be a big role for people. And most of my clients at least, even though they use AI as a tool to assist them, they’re still very big believers in the importance of the human contribution.

Orlando (59:20.141)
Yeah, and I think that people forget that that creativity, that creating a movie is fun. It’s a form of entertainment itself. And I think that that’s that’s gonna become much more a part of I think the discussion. I mean, Roblox has, you know, hundreds of millions of users. at least 50%, if not more, are the creators of the Roblox games. It’s kids going on there to make a game that other people will play.

Lisa Oratz (59:49.075)
Mm-hmm.

Orlando (59:49.483)
And that’s entertainment too. That’s not just a monetized engagement. That’s fun.

Lisa Oratz (59:52.083)
We shifted yeah, we shifted from this notion of entertainment as a passive thing to as a something as a participatory thing and I think that’s continuing. I don’t think that’s you know, I don’t think that’s going away. I think it’s a good thing. You know, I think many people can think that that’s you know, that that’s a good thing.

Orlando (01:00:06.572)
Yeah.

Orlando (01:00:14.859)
Yeah. Well, I mean, Lisa, thank you for that. On that note, I mean, I think that’s a great place to end. Thank you so much for doing this with me. I’m I’m, you know, I’m I’m always completely fascinated when you talk about all this. And you know, and I’m I so I’m so grateful that you came on the podcast and and you know, I’m I’m still just somebody learning on the other side of it. And I so appreciate how you talk about these things and and you know, and and the education certainly that I’ve gotten from you.

Lisa Oratz (01:00:43.133)
Well thanks for having me. I love talking about it. It’s fun, it’s entertaining and you know, I really do want to help get some good information out there because there’s just so much confusion. So thank you for for letting me share a little bit.

Orlando (01:00:53.211)
it’s a pleasure. Well, maybe when some of these other decisions come come out, we’ll we’ll have you back. All right, terrific. Thank you so much.

Lisa Oratz (01:00:58.171)
Okay. Happy anytime.

You’re welcome. Take care.