Navigating AI and Copyright Law Insights from Legal Expert

**(00:55)** If this is your first time joining us, welcome! These discussions are interactive, and we encourage questions, especially today with such a critical topic. Brooke will focus primarily on copyright issues related to AI. I'll let her introduce herself and share how she got interested in AI and why it has become a significant focus for her.

**(02:11)** Brooke: Hello, everyone. It’s nice to see so many familiar and new faces here. I’ve been working with tech startups for a long time, which naturally led me to work with clients involved in AI. As an attorney, I follow where my clients' needs go, and for many, that’s increasingly into the AI space. I’m a solo practitioner, though I do have an assistant who helps keep track of AI updates—they’re coming fast and furious these days! Just this morning, I got an alert about new U.S. initiatives to maintain leadership in AI, showing just how rapid these changes are.

**(02:43)** I have a presentation prepared, so I’ll share my screen. [Shares screen] And as David mentioned, this can be very interactive, so feel free to ask questions along the way. If I don’t see you, just speak up.

**(03:17)** To start, let’s take a high-level look at ownership and intellectual property. Property generally falls into two categories: real property and intellectual property, and property ownership exists because of laws defining and enforcing it. Intellectual property (IP) includes patents, trademarks, and copyrights. We’ll focus on copyright today, covering what rights owners have, what others can do with those rights, and how copyright specifically applies to AI.

**(04:27)** Copyright law protects original works, covering a broad range like computer code, audio recordings, video, photography, artwork, books, and blogs. Copyright law is federal in the U.S., unlike privacy laws, which are often a state-by-state patchwork. The U.S. Code Title 17 governs copyright, with oversight from the Copyright Office.

**(04:57)** Ownership of a work usually belongs to the creator—whoever "puts pen to paper" or "fingers to keyboard"—unless there’s an agreement stating otherwise. Many companies require employees or contractors to sign agreements specifying that any work they create belongs to the company. It’s always best to establish ownership rights upfront, even though ownership technically exists upon creation without registration. Registering a copyright, however, offers advantages if you need to enforce it, such as access to statutory damages.

**(06:02)** For copyright protection to apply, two conditions must be met: the work must be original, and the creator must be human. The originality requirement means the work can’t simply replicate something that already exists, though it can be a derivative work. For instance, a movie based on a book is a derivative work, as are new editions of textbooks and translated texts. It must incorporate the pre-existing work while adding new, original authorship.

**(07:03)** As for the human requirement, it may seem obvious, but AI complicates things. The Copyright Office has rejected copyright claims for works claimed to be authored by divine or non-human entities.

**(07:36)** Here’s an example: a photographer set up a camera, and a curious monkey took a selfie. The photographer attempted to claim copyright, arguing he set up the conditions, but the Copyright Office ruled that only humans can hold copyright. Even an attempt by PETA to claim copyright on behalf of the monkey failed. 

**(08:45)** Now, onto AI-generated content. The Copyright Office issued guidance in March 2023, addressing whether works are “human” if they involve AI. They assess whether the human created the work, with the computer as an assisting tool, or if the machine itself generated the creative elements. When users simply prompt an AI, they’re not considered the author, as they lack ultimate control over the creative output—similar to commissioning an artist. However, if someone modifies AI output substantially, those modifications may be protected.

**(09:59)** Any questions on that so far?

**(10:30)** David: I have one: what about scenarios with tools like Photoshop that might have automated features? 

**(10:30)** Brooke: Great question. It’s always a case-by-case analysis. Think of it as a winding river rather than a strict line. Where creative control shifts from human to machine is still being debated. The Copyright Office is actively seeking feedback on whether the current approach to AI-generated works is appropriate.

**(11:36)** Attendee: If you prompt an AI multiple times to refine an output, is that considered human input?

**(11:36)** Brooke: That’s a gray area, and there’s no official guidance on it yet. A relevant example is “Zarya of the Dawn,” a graphic novel created with over 400 AI prompts. Despite the extensive prompt work, the Copyright Office denied protection for the images but allowed it for the text. Many, myself included, think they got it wrong there, as extensive prompting arguably shows creative direction.

**(12:51)** For example, using AI to blur elements in a “man-on-the-street” video as part of your editing doesn’t impact copyright eligibility; it’s considered a tool. But if you use AI-generated special effects, the effects might not be protected. Translations by AI also don’t qualify for copyright since human creativity is involved in literary translation decisions. It all comes down to whether AI is enhancing or creating independently.

**(15:44)** Attendee: How does the copyright registration process work?

**(15:44)** Brooke: Briefly, the moment you create a work, copyright exists, though registration provides benefits for enforcement. The Copyright Office has forms, videos, and resources to guide you through registration.

**(16:13)** Now, once you own a copyrighted work, others can only use it through permission (like licensing) or under fair use. Licensing includes music, templates, photos, software, and more. Fair use, however, balances the copyright holder's rights with the public interest to encourage creativity and innovation. It’s assessed with a four-part test examining purpose, nature, amount, and market impact. This test is challenging and can lead to differing interpretations, even in the courts.

**(18:28)** David: That’s helpful context, especially with AI tools like Augi Studio that can generate voiceovers, scripts, and video montages from stock clips and your own content. It seems like if there's minimal human input, claiming copyright for the final product would be unlikely.

**(19:02)** Brooke: Correct. Using only AI-generated scripts and stock footage wouldn’t likely qualify for copyright.

**(20:05)** Here’s a timely example: in *Andy Warhol Foundation v. Lynn Goldsmith*, the Supreme Court ruled in favor of the original photographer, Lynn Goldsmith, whose photo was used by Andy Warhol in a piece licensed to *Vogue*. Goldsmith argued that Warhol’s use harmed her market, as she could have licensed her photo to *Vogue* herself. This ruling prioritizes protecting the original work, which is particularly interesting now as AI models use copyrighted materials for training.

**(23:47)** In the AI landscape, some companies are paying for licenses, like Reddit, Google, OpenAI, and Meta, while others rely on fair use, especially smaller companies without the funds to license materials for training. Fair use here depends on factors like whether the AI outputs are creative enough to be transformative or if they directly compete with the original.

**(25:27)** Attendee: So, how should we approach AI use with copyrighted material?

**(25:27)** Brooke: Good question. When using copyrighted material, consider how important ownership is for your project, the potential infringement risk, and the protection level you need. For example, if you’re creating AI-generated code for a proprietary tech product, investors will want assurance that you own it. Use a risk calculation, considering both likelihood and magnitude of potential harm.

**(28:16)** Attendee: As a brand not building AI tools but using AI solutions, should we worry about AI-created materials?

**(28:16)** Brooke: Yes, especially if clients or customers are wary of AI-created content. Ownership and potential infringement risks depend on how crucial originality is for your project and how you’ll use the outputs.

**(29:27)** Moving on to publicity rights: individuals control the use of their image, likeness, and voice, mainly regulated by state laws. California has the most protective laws in this area, while Nevada’s are more relaxed. For example, in *Kim Kardashian v. Old Navy*, Kardashian sued Old Navy for using a look-alike in an ad, and they eventually settled.

**(33:24)** David: So if we use a Robert De Niro-sounding voice from an AI tool like Eleven Labs, is that okay?

**(33:24)** Brooke: Generally, no. Even if it “sounds like” De Niro without explicitly being him, if it implies it’s him, you’re in risky territory.

**(36:48)** Now, where are we headed? In October 2023, the Copyright Office issued a notice seeking public comments on four key topics: whether using copyrighted works to train AI should be considered infringement,

 how to handle AI-generated works, liability for infringing works produced by AI, and considerations for AI-generated likenesses. These are complex issues, and the results could influence future legislative or judicial decisions.

**(39:14)** The White House also issued an executive order in 2023, with agencies assessing AI’s impact on various sectors, including cybersecurity, a top concern given potential misuse by bad actors. The EU’s AI Act takes a similar approach, focusing on responsible AI use.

**(39:53)** Attendee: Thanks, Brooke, for this in-depth discussion. I have one last question on building authorized AI models that embed provenance into outputs. How might that impact copyright?

**(40:54)** Brooke: Interesting. Provenance-tracking adds value but doesn’t necessarily create copyright where it doesn’t already exist. That said, provenance could be helpful for licensing, and it’s likely we’ll see AI-driven provenance become part of legal structures around AI content.

**(44:43)** Provenance tracking is a good direction, and any copyright owner who can trace their work would benefit from it. In the past, broadcast flags were used to trace video, particularly by networks like NBC for events like the Olympics.

**(45:22)** Attendee: In California, marketers often negotiate contracts with influencers. Could companies legally use such contracts to create AI avatars of influencers?

**(45:47)** Brooke: It depends on the contract. Sometimes, releases are very broad, giving the company extensive rights to likeness and image across future uses. As an attorney, I’d advise narrowing it to specific uses. Celebrities, for example, often sign extremely limited agreements about how their image can be used.

**(48:05)** Paul: I used to run a comedy company, and parody was critical to our work. With deepfakes, Elon Musk has argued they count as parody. Is that true?

**(48:40)** Brooke: Not inherently. Parody applies to copyright, not to likeness rights. For a deepfake to qualify as parody, it would have to comment on the original work in some way.

**(49:54)** David: Thanks, Brooke, for such an insightful discussion! We’ll wrap up here, but I hope we can have you back for more, as I know this only scratches the surface. For everyone, we have more exciting sessions lined up, including a talk with Huge’s CTO and our first APAC event next month. Thanks, everyone, for joining today, and enjoy the rest of your week!