And like most legal questions, the answer to “Can I use AI to generate work for clients?” is, “It depends.” We’re in such a gray area with AI right now that the legal system hasn’t had a chance to catch up to technological advances. So how do agencies protect themselves, their workers, clients, and their IP when everything is so up in the air right now?
Sharon Toerek has some common sense advice for approaching AI in the workplace to avoid a potential legal nightmare. In this episode, we’ll cover human-generated vs. AI-generated content, the gray areas we need to look out for when using AI, how to protect client and agency copyrights, how to avoid infringing other copyrighted content, and so much more.
AI-generated content is an incredibly important and ever-changing topic. And if you’re an agency owner, this needs to be addressed in your agency today.
A big thank you to our podcast’s presenting sponsor, White Label IQ. They’re an amazing resource for agencies who want to outsource their design, dev, or PPC work at wholesale prices. Check out their special offer (10 free hours!) for podcast listeners here.
What You Will Learn in This Episode:
- The top 3 questions lawyers are receiving about AI
- Having conversations with clients about the use of AI-generated content
- The gray area between AI-generated and human-generated content
- The ownability of AI-generated output
- Creating documentation about implementing AI tools in a typical workflow
- Including AI prompt strategies in trade secrets documentation
- Can we protect the data we’re inputting into the collective that is AI?
- Addressing the use of AI in SOWs
- Should we use AI tools to create legal documents?
“I have always said that technology leads business and sort of runs behind it with its sneakers on, and then here come us lawyers sweeping up the mess behind all of that.” @SharonToerek Click To Tweet
“Anything generated from an AI tool, according to the copyright office right now, is not ownable in terms of copyright. Only human-generated content is entitled to copyright protection.” @SharonToerek Click To Tweet
“One of the central hallmark questions to consider between agency and client is: how ownable does anyone care about the work product being?” @SharonToerek Click To Tweet
“Always include attribution if you're going to use an AI-generated image or graphic so that the viewer is very clear on the fact that a machine generated this.” @SharonToerek Click To Tweet
“I think there's opportunity to look at your prompt strategies and the uniqueness thereof and protect them under the theory of trade secrets because they're not copyrightable.” @SharonToerek Click To Tweet
Ways to contact Sharon:
- Website: legalandcreative.com
- LinkedIn: https://www.linkedin.com/in/sharontoerek/
- Twitter: https://twitter.com/SharonToerek
- The Innovative Agency Podcast
- My Future Self: https://agencymanagementinstitute.com/myfutureself/
- AI and Agencies: https://agencymanagementinstitute.com/jkfiwoiduaosa/
It doesn’t matter what kind of an agency you run, traditional, digital, media buying, web dev, PR, whatever your focus, you still need to run a profitable business. The Build A Better Agency Podcast, presented by White Label IQ, will show you how to make more money and keep more of what you make. Let us help you build an agency that is sustainable, scalable and, if you want down the road, sellable. Bringing his 25 plus years of experience as both an agency owner and agency consultant, please welcome your host, Drew McLellan.
Hey, everybody. Drew McLellan here from Agency Management Institute, coming back to you again with another episode of Build A Better Agency. Super excited to have our guest back with us today, and we’re going to talk about something that I know is on your mind because I’m talking with you about it on a regular basis. She just has an area of expertise around it that I do not. So hence, my invitation to her to join us.
So many of you are familiar with Sharon Toerek. She is an attorney who specializes in her firm, specializes in working with agencies. So many of you have engaged them with intellectual property filings. You have used them for master services agreements. You are partnering with them to have review contracts that clients give you. One of Sharon’s areas of expertise, as I said, is intellectual property, trademark, copyright, all of that sort of thing.
So with all things AI happening in our world, I knew that we needed to get her on the show to talk about, from a legal perspective, what’s going on with AI and agencies, what can we protect, what can’t we protect, what kind of conversations should we be having with clients, what kind of documentation do we need to have, recognizing that whatever we have to have today, in six months it’s going to be something different, but we’ve got to try and stay current with how we understand how AI is impacting our business from a legal perspective, the risks, the opportunities, and what we can do to protect ourselves and our clients.
So I don’t need to belabor that point anymore. I know you’re feeling the pressure of understanding these things. So I would like very much just to get to the conversation because I guarantee you an hour is not going to be enough time for me to ask Sharon all the questions that I have. So let’s get to it.
Sharon, welcome back to the podcast.
Thanks, Drew. I’m excited to talk with you today.
So it’s been a little while. So normally, when you’re here, we’re talking about master agreements or partnership agreements or some of those things, but today, I want to be really specific, and I know you and your firm are doing a ton of work around, as always, intellectual property, but AI changes the game completely. We get a lot of questions from agency owners saying, “What do we own? What do we not own? How do we protect who owns this stuff if we’re creating it in this new space?” So I just knew that it was time to get you back on the show and really to talk about specifically some of the challenges around AI and deliverables and work parts and pieces and all of that. So just give us a little overview of the kinds of things that you’re talking to agency owners about in terms of AI, and then I have a host of questions for you.
Yeah. Well, every time you think the world cannot get more fascinating for the agency world, for the agency industry, it gets even more fascinating. I think what’s the most fascinating about AI and its impact on agencies right now from a legal perspective is not that we didn’t see something coming, but the speed with which the technology is improving and changing. I have always said that technology leads business, runs behind it with its sneakers on, and then here come us lawyers sweeping up the mess behind all of that. Then this is a perfect example of that, but running at warp speed around the track.
So we are starting to get quite a few questions, and they center around I think a few areas. The first is wanting clarity around the ownership stake that an agency, and by extension clients, will have in any IP that is generated using AI tools.
I would say the second is the conversations that you need to be having as an agency with your client before the work is generated. So to create reasonable expectations about output and ownership and all the other things that go along with it. Then third is what are the opportunities for agencies in terms of how they’re implementing AI into their workflow. Is there anything proprietary or protectable around that? If so, they’d be thinking about.
I would say the conversations are probably the most critical thing right now between agencies and clients because that’s what’s going to set the expectation for the relationship moving forward.
Yeah. So we’ll dig into that. So I’m hearing a lot of agency owners and leaders talking about what tools are people using, what are the implications, but they are worried about the conversations with clients. Do they disclose that they’re using AI? Does it change ownership and trademark and IP registrations and things like that? How do they deal with the pricing aspect of it? There’s just a lot of nuance in those conversations. So when you talk to your clients about how they should be having conversation, what are you recommending that we do around these conversations?
Yeah. Well, I think the first thing is develop a point of view around what AI generative tools you plan to use to fulfill the engagement and have a conversation with the client about that and be ready. I mean, in some cases, you don’t have a choice because they’re already asking you what percentage of the output is going to be generated by AI, what are you going to do to the AI generated output, if anything, how much human contribution to the work product is going to be, those sorts of things. So you need to start to … You can’t answer them with 100% definiteness because the world is changing so quickly, but you can know, for example, that approximately X percentage of the output or these specific projects are where you’re going to explore the use of AI-generated content or output.
For an example, let’s say we’re going to do a branding campaign for you, we’re to support the branding campaign with perhaps some original blog content and social media posting, community management and a few other things. These parts of it, the blog rating, the social media post generation, and maybe some of the artwork that supports those things, we’re going to experiment with AI tools because we think it can help us speed up the output and we think we have a good handle on how to polish that stuff up so that we can put it out there so it’s a good representation of your brand.
So those are the areas where we’re going to use the AI tools. Your brand development, naming, logo development, and research, we may look at AI just for inspiration, but we’re going to create original brands for you, and we’re going to go through normal channels to do our due diligence legally around the availability of any new branding that we recommend to you, and so on and so on and so on. So I think that’s the first part of the conversation is, “Here’s where we’re thinking AI is going to be additive to the process and why.”
So after that, a couple of things. Client need to understand about any AI-generated output. Anything that spits out of an AI tool is according to the copyright office right now, not ownable in terms of copyright. So only human-generated content is entitled to copyright protection. What does that mean? It may mean that your prompting strategies are absolutely protectable, but not by copyright law, probably by trade secret law, and only dependent upon the extent to which you keep them confidential and only share them either with the client or internally in the agency. The output, if it comes out in the machine and is untouched, not protectable under copyright, under trade secret, under most theories of IP protection right now.
What about the stuff that you plan to then put into human hands to spiff up to build upon? That’s an open question right now regarding … Where I come down on it and where most lawyers I talk with come down on it is to the extent you can draw a bright line between what humans have created and what AI has generated, you could own the human created components, just not the AI-generated components.
So let’s talk about that a little more specifically. So I go into ChatGPT or some other tool and I say, “Write me a blog post in the style of these three blog posts about Drew’s opinion to how to sell your agency,” and the AI and ChatGPT gives me a blog post that is in that same style and I go in and edit it. So I use that as my first draft, if you will, and I go in and edit it. Is there a level of editing that makes it ownable or protectable? Is it a percentage of changes or word count changes? Is it just if I make minor edits, it’s not owned or protectable by anyone? I mean, how do I know what enough editing or polishing gets me to a point where I can own it?
Yeah. There’s no good answer to that question right now, unfortunately, in full transparency, and I would like to even be able to say it depends, the lawyer’s favorite phrase, “It depends on this. It depends on that.”
I was going to say, normally, you at least go with it. It depends.
Here’s why I say that. This is not the first conversation I’ve had about this.
It depends upon the extent to which you can separate the human contribution from the AI-generated content. When you’re talking about editing or you’re talking about minor modifications, there’s no percentage, there’s no number right now. Again, this is going to iterate quickly and case law will occur, but these things move slowly. Regulatory change moves slowly. Well, the copyright office has moved more quickly on this to make some sort of a statement on the protectability at all of AI-generated content than any practitioner has seen in a super long time. So I’m impressed that they even settled that point.
So yeah, there is no clear way to distinguish. The only clear ways to distinguish would be the original AI-generated content. The blog that we used for our blog post is probably not that ownable, if you will, but the attribution to you and your thought leadership and the links to the other content that you’ve created, of course, all that’s ownable. You might have added an image to it that’s protectable under copyright law.
A question I get a lot, by the way, related to this is, “Well, what if you want to own and protect the human-generated content that sits alongside it in a finished work?” You can own that. That’s still ownable under copyright law because it’s human-created and you can identify the creator. It’s just this gray area regarding the ownership of the AI-generated content. Unfortunately, no guiding lines or even sliding scales to tell us how much do you have to change it before you can say, “I own it.”
So the only other unofficial guideline I would give you is the extent to which it is spitting out stuff that is repetitive of or iterative of other content that’s out there that was human-created, the better chance you might have of claiming it as proprietary to you.
So I know a lot of agencies, they’re not just doing this with words. A lot of agencies are playing around with images and the output is remarkable. They’re going into tools and they’re saying, “I want to create an underwater scene that I’m going to rest a pearl on,” and yada, yada, yada or in some cases, I saw an agency that went in and asked for some packaging design recommendations, and I swear to God, in 10 minutes, they had about 300 different … They gave it tone and style and color palette. So in a lot cases, they’re using that as, again, first draft inspiration, and then they’re modifying that design or graphic or photograph or whatever the output is, but again, they are starting, the core product they’re starting with is AI-generated.
Yeah. I think that’s a terrific example, and here’s why. That is an example of something where the end client would very clearly care about the ownability or a level of safety from infringement versus you’re going to write four original blog posts for us a month and they’re going to be topical and they’re going to be cycled and fall lower and lower and lower down Google over time, which we care less about. So one of the central hallmark questions to consider between agency and client is how ownable, does anybody care about the work product being your brand. I do not recommend using AI-generated tools to do your due diligence on the availability of trademarks, for example.
Oh, for gosh, right.
That’s today as we sit here in the first half of 2023. I don’t know if I’ll be making that recommendation a year from now. I assume somebody’s at work on a trademark searching tool that is more reliable, but now, you got to still do your due diligence. So that’s an example where there’s an extreme level of care about the ownability versus areas where not so much. So assess that right as an agency.
Start with the end in mind on that.
Do we need to be able to defend and protect this?
Right, and what’s the lifespan, unintended reach and distribution of the completed work product, in what markets and for how long, and how frequently are you going to redo this work. All those factors go into the central question of how much do we really need to care about how ownable this IP is or are we doing this for a brief moment in time, historically speaking, for our business. So I think those are important questions.
My experience working with agencies, particularly in the contract negotiation phase of the relationship with their clients, is they’re not super big fans of vulnerability when it comes to the legal conversation. This is one area where I think I’m adding this to my list of two. This is one of those areas where you have to get over yourself and really go in there and chat about this honestly and with vulnerability and transparency, just as I highly recommend you get over yourself about not wanting to sign a nondisclosure agreement with a potential new client, which is a topic we can talk about on another day, but this is the second one where you really have to sit down and say, “Look, let’s talk about the end in mind,” as you said, Drew, “and then let’s talk about how we’re going to use AI-generative tools, which ones we think we might be using, and where the likelihood of AI-generated content might emerge, and let’s talk about whether that’s a problem for you or not.”
Enterprises are maybe going to be a little less understanding about this. They’re going to want agencies probably to take more of the heat on the liability for infringement, but that’s a negotiation point in your contract. Don’t be afraid.
I would think the waters are even more muddy when it comes to images than they are with words. At least words you can count and you can say, “Well, that used to say blue and I changed it to cow,” or whatever the thing is, but with images, I would just think the nuance of making changes, how much shading did you add, did you change a color, did you move the shape a little bit. It would be much harder to defend, argue, quantify. So I think that’s going to be even a more sticky wicked, I think, for folks moving forward.
Right. I agree with you, and I think your best guideposts there are really the elements, the good old-fashioned elements of copyright infringement. I mean, quickly to establish copyright infringement, there’s two big things you have to prove. You’ve got to prove access to the original underlying work that you’re saying got infringed and you have to prove substantial similarity. So there’s a whole list of factors that go into determining whether two works are substantially similar. How much of the original did you take? Did you take the most important part of the original? So on and so on and so on. So I think it’s the same analysis here.
First of all, always, always include attribution if you’re going to use an AI-generated image or graphic so that the viewer is very clear on the fact that this was generated by machine. Then secondly, if someone is saying that yours looks like an infringement of theirs, then you have to quickly evaluate how similar are they, and how much of the original work is reflected in the generated work, and then make some quick decisions. Fortunately, although I’m certain they’ll do it one of these days, Getty hasn’t yet started, they’re creating AI images, but I don’t think they’re necessarily going to enforce licensing at least right now the same way that they do for human-created images.
So crazy. So is part of what you would use to prove either your originality or potentially I guess the lack thereof the prompts you used to create the piece?
I think so. I think that as much as you need to think about the protectability of your prompt strategies, and I am convinced that this will become a core service related to AI that agencies could excel at over time would be, how do you develop your prompt strategies to get the best possible output out of AI? Agencies are really great about thinking about these problems this way, in my view.
So as much as you need to think about the protectability and the wisdom of protection of your prompt strategies, it’s documenting them so that if you’re ever called on the carpet for an unintended infringement, you can document, show your documented prompt strategies to say, “We had no way of knowing that the output came from …” The one thing you got on your side is you don’t know where it came from exactly because the AI technologies are ingesting massive amounts of data, and you’ll have no way of knowing specifically, at least for a while, as a user of the tool what got fed into it and therefore what might show up in the output.
Right, but I think being able even to show what the prompts were and what the initial output was compared to the final would at least be able to show iteration and intention maybe of-
Correct. I mean, and if you’re in the unfortunate position of having fed it an original image and the output looks startlingly like that, that’s not going to be good for you. So be smart about that, and if your output is not distinguishable enough from the original, you get some advice about that from council or whomever and try again. I mean, that’s supposed to be the beauty of using these tools is that they’re fast and inexpensive. So try again. I mean, it’s going to be some art and craft in addition to the science and the tech here in terms of using these in a way that generates something that’s original enough to be effective in marketing, first of all, but protectable from an IP perspective.
So you had mentioned that if copyright isn’t our friend in terms of protecting these things, there are some other legal aspects that we could potentially use. I think you said trade secrets, for example. So talk to us about how one might be able to use other things other than copyright to protect some of the work that is created in partnership with AI tools.
Right. So part of the framework that we use at the firm to assess IP and how and when you should protect it, and the strategies to do that, it’s a triangle analysis we go through. One point of the triangle is your content. That’s the best equation here to what we’re talking about. So your content is either protectable under copyright, which we’ve already pushed that to the side for the time being. Trade secrets are a way to think about protecting your input into the AI generative machine. To the extent that you have developed a process for determining what your prompts are going to look like to the extent that you have unique industry or skill levels in your agency in an area that a lot of competitors don’t, and therefore your prompts will reflect that fact, I think there is some protectability under the theory of trade secrets, but they require a couple of things.
You legitimately have to keep them confidential, which means that you’ve got to have a contract with your client that speaks to that confidentiality, and you have to have the conversation with your client about whether the agency is going to own those prompts or whether the client actually owns the prompts, provide them to other agencies or other internal people to generate other work product together or with other prompts.
So I think that there’s opportunity in looking at your prompt strategies and the uniqueness thereof and protecting them under the theory of trade secrets because they’re not copyrightable. There’s not enough, I don’t think, originality in them, and you don’t want to disclose them, which you have to do to protect under copyrights. So that’s, I think, a strategy that a lot of agencies need to think about, and it needs to be reflected in the agreement you have with your client, whether it’s an MSA or in the SOW. That’s an open question. I don’t care which one it’s in as long as it’s in one of them that’s signed by everybody.
All right. So I’m going to ask one more question, then we’re going to take a break, and then we’re going to come back and just talk about things like MSAs and other documentation, but my question is, okay, great. So I have a trade secret or I have some thought leadership that is unique to our perspective and how we do the work, so in theory might be protectable in that way, but I’m putting it in a tool that makes it very clear that the tool I’m putting it in will then use it in other … It becomes part of the collective wisdom, right?
So it’s not like ChatGPT or any of the other tools are keeping what I put in as prompts. They’re not promising me privacy or exclusivity or confidentiality in that. So aren’t I putting my trade secret into something that is not protectable? I can’t say to my client, “Hey, no one at ChatGPT or the machine itself is ever going to see this, use it, do anything with it.” I can’t really do that, can I?
No, you can’t, and there’s a ton of conversation occurring right now around the protectability of and confidentiality of data that’s input into an AI tool. You can’t protect it forever. I think you need to think about this in terms of an opportunity to develop a proprietary process that develops your prompt strategies. What are the factors that go into? Once the prompt strategy is fed into the machine, it becomes part of the collected wisdom, but I feel like the likelihood of that exact prompt strategy being used in a way that would diminish the value of your proprietary process for generating is still pretty low. That may not be true a year or five years from now, but we’d have to grab the corners of IP that we can protect right now. I think I still feel like this is one area of real estate that could present an opportunity.
Okay. All right. Let’s take a quick break and then I do want to talk about what do we need to do in the short term just to manage the legal implications of all of this as it is changing every freaking day. So let’s take a quick break and then we’ll come back and talk about that.
Hey, just want to take a quick minute and tell you about a resource that we have on the website that I don’t talk about as often as I should. So it’s an exercise called My Future Self, and the reason why you would do this exercise is if you are in planning mode, and this is really for you as either an agency owner or an agency leader, but you really want to think about what your future looks like, not the agency’s future, your future.
I find so many agency owners struggle with how they are spending their days and is it fulfilling and is this what they want to do in five years that we created an exercise, and I will tell you a very brief story, but I first did a version of this exercise probably 15 years ago, and it basically walks you through some thinking, and you have to do some journaling around what your future self looks like. You have to give yourself into it. You have to really suspend the reality and talk about what it is today, but I’m telling you when I did it, how it was different from my current moment in my life was pretty dramatic.
I was working with a coach at the time and I said, “This is great, and this is the life I want, but it doesn’t look like my life now.” We talked about just being open to the possibility of transitioning in some of those directions. I will tell you, for the last 10 years, I have been living that life, the life that I created in this exercise.
So it can be very powerful and very eye-opening. I’m not a woo-woo kind of guy, but once I understood what I wanted as opportunities presented themselves, I just took advantage of them in different ways than I would’ve had I not done this exercise. So head over to the AMI website and go to agencymanagementinstitute.com/myfutureself. You can read more about it. There’s an intro, a video intro where I tell you all about it and then some questions. It’s $197. If you don’t like it or you don’t end up doing it, happy to give you your money back, but I’m telling you, it can be really transformative if you give yourself into the exercise and really do it with an open heart. So just wanted to tell you it was there. Hopefully it’ll be helpful for some of you. Okay. Let’s get back to the show.
All right. We are back with Sharon Toerek and we were talking about the legalities around AI-generated work. So talked a little bit about where things are at today, but I would like to think that agencies are feeling like they need to cover themselves in some documentation or ways that they, even though it’s an ever shifting landscape, that there are some things that they need to do. So let’s talk about things like scope of work documents, MSAs, that sort of thing. What should we be sure to include in those kind of documents?
Right. So I think that every MSA needs to address ownership, clearly address ownership and timing of transfer of IP, and it also-
Which has always been a rule. It’s always been a thing, right? This is not a AI thing.
Right. You and I have known each other, but you’ve probably heard me say it 200 times, easy, and it always has been, and there’s always language that we recommend in that section of any MSA that deals with third party generated IP. So for example, in the olden days, six months ago, in the before times, there was always we need to use some stock photography. We need to license some font. We need to license software code to include in what we’re going to program for you, et cetera, et cetera. We don’t own that. It’s owned by a third party. We’re going to get a license. You’re only going to have the rights that we have.
Same concept really adapted specifically to AI-generated content or output. That’s a section of your agreement. First of all, you need that section. So if your MSA doesn’t have a section addressing third party created intellectual property, it needs one, and that section should now include, I believe, a mention of AI-created stuff, whether it’s content, imagery, whatever it might be. You need to include a representation that the agency may use AI generative tools to create some of the material that ends up in the deliverables, the client acknowledges this, the client acknowledges that neither the agency nor the client will own anything that’s completely machine-generated, et cetera.
So really, this is just a bullet point in the … Sometimes we go to outside sources to generate bits and parts and pieces of our work. So again, it is everything from photos and fonts to code, to contractors and now AI just falls in that list.
Correct, and that’s one place where you need to mention it, I believe. I think you also need to mention it in your warranties and in your indemnifications. So you can’t warrant the originality of something that AI generates because you don’t know if the data that got fed into the tool was accurate or free of any infringement in the first place. So you certainly don’t want to be in the position of warranting that to your client.
So in the section of the MSA that discusses, and trump representations, warranties, indemnificati