Decoding AIs Legal Landscape Gary Kibel on Copyright Creativity and AI

(00:49) When I received one of these updates, I immediately reached out to Gary and asked if he’d help us make sense of AI's legal landscape. So often, these discussions raise legal questions, but lawyers aren’t in the room right away. Gary, it's great to have you here—we’re looking forward to hearing your insights.

(01:23) **Gary Kibel**: Thanks very much, David. Happy to be here, and I recognize a few names here on Zoom. I’ll start by sharing my perspective on legal issues in the AI space for about five minutes, then we can open it up for discussion. As David mentioned, I’ve been in advertising and marketing law for over two decades with Davis+Gilbert, a firm deeply involved in this space. While I lead the privacy and ad tech practice, AI is a huge part of what we’re dealing with today. 

(01:57) AI raises several legal concerns—primarily around training data, input data, and output data. Legally, we see issues in copyright, trademark, right of publicity, contracts, and privacy. Starting with copyright, when someone creates a work of authorship—say, writing a book, painting a picture, or taking a photo—they own the copyright. But with generative AI, who owns the output?

(03:06) If you read OpenAI’s terms and conditions, they essentially say, “You own the output, but we can’t guarantee ownership beyond that.” The U.S. Copyright Office recently made a statement that a computer cannot own a copyright—copyright belongs to original work created by a human. This isn’t a new concept. For example, Google “monkey selfie” to see a case from about a decade ago, where a photographer’s camera was used by a monkey to take a selfie. The photographer tried to copyright it, but the office denied it, saying the monkey, not he, took the photo. This question of ownership also applies to AI-generated output.

(04:23) This issue extends to trademarks and contracts as well. For instance, if an ad agency uses generative AI to create content for a client, the work may not qualify as “original,” which could violate contractual terms if originality and ownership were agreed upon. Some agencies are now including AI clauses in contracts to clarify ownership issues.

(05:59) Similar issues come up with the right of publicity, especially with deepfake likenesses of public figures. We’re seeing lawsuits, such as the one from Sarah Silverman against OpenAI, arguing that AI platforms are scraping copyrighted material as training data without permission. Getty Images has also sued AI platforms, assuming they’re scraping their copyrighted images. 

(07:06) The main advice I give clients is to be cautious with both input and output: don’t request AI to generate something highly specific that could infringe, and carefully review the output for any potential issues. For instance, asking for a song in the style of a well-known artist like Taylor Swift could lead to trouble, whereas asking for a generic happy song would be safer.

(07:33) **David**: This is all great. So, what are lawmakers and regulators doing about AI?

(07:42) **Gary**: In the U.S., we don’t yet have a specific AI law. Privacy laws are still evolving, and now regulators are turning attention to AI. The Federal Trade Commission (FTC) has issued statements around AI-related deception—both for companies using AI without disclosing it and for those claiming AI involvement when it’s just for hype. Actors, represented by unions like SAG-AFTRA, are also concerned about AI replacing them. In Europe, Italy even banned ChatGPT temporarily due to privacy concerns, though that ban was later lifted after OpenAI made changes.

(09:16) **David**: This is why I feel like we could have you here weekly and only scratch the surface. Regarding agency contracts, are you seeing new clauses added for AI use in contracts with clients?

(09:54) **Gary**: Yes, definitely. Recently, I saw a clause in a contract from a major company, stating that the agency couldn’t use generative AI without the client’s approval. It seems they wanted to address AI use upfront to avoid any issues.

(10:32) **David**: I wonder how enforceable that is, especially with remote work. How can agencies ensure compliance when work could be done privately, potentially assisted by AI?

(11:09) **Gary**: It’s a challenge, particularly with “inspiration” being a gray area. Some companies and law firms have internal policies on AI usage. There’s a high-profile case where a lawyer submitted a brief generated by AI that cited fake cases, which resulted in disciplinary action. It’s one thing to use AI for brainstorming; it’s another if AI-generated work becomes the deliverable without transparency.

(13:15) **David**: For brands using AI, are there contracts or rules for inputting proprietary brand data into private AI models, like one specific to Coca-Cola’s brand guidelines?

(13:58) **Gary**: Yes, we’re seeing more closed environments where companies input their proprietary data to control the training data, like Adobe’s Firefly and LexisNexis’s AI product. These setups ensure the training data is proprietary and safer for use.

(15:42) **Paul**: Regarding ethics, what’s the responsibility balance between AI developers, marketers, and agencies?

(16:13) **Gary**: The platform developers, like OpenAI, have terms to clarify who “owns” the AI-generated content, but they don’t ensure it’s original because they use vast datasets as training data. The lawsuits claim these platforms must be copying something since they produce content that closely resembles existing material. Copyright law, however, protects original creations, not ideas or general inspiration.

(18:17) **Brooke**: How is AI different from a designer or copywriter drawing from years of experience? AI also pulls from vast amounts of data, but it’s now considered infringement if AI-generated content resembles something existing.

(19:12) **Gary**: That’s a fair point. Human creators draw from experiences, but they typically don’t remember every source, which makes infringement less likely. With AI, you don’t know where the output is derived from, which makes it riskier. Closed AI environments attempt to control this by using specific, vetted data.

(20:59) **David**: With AI increasingly integrated into tools like Google or Dropbox, is it possible to know when AI is in use?

(21:38) **Gary**: I would advise agencies to view AI as a tool for inspiration, not as a direct deliverable. Some companies allow AI for brainstorming but don’t let it be the end product without thorough review.

(22:15) **Paul**: On copyright—since computers can’t own copyrights, at what point does ownership shift to humans?

(22:48) **Gary**: Good question. The Copyright Office suggests that if you register AI-assisted work, you must specify the AI and human-created parts. Humans retain rights to the original, human-created parts, but it’s a gray area that will likely be clarified through litigation.

(24:29) **Paul**: If everything becomes a mix of AI and human content, does copyright need redefining?

(25:02) **Gary**: Copyright protects creators by granting them exclusivity to their creations, unlike trademark law, which protects consumers. How we define ownership of AI-assisted content is evolving, and upcoming cases will likely provide clarity.

(26:27) **Brooke**: A piece of art won a competition, but AI was integral in generating it. The Copyright Office ruled it wasn’t copyrightable because of AI’s involvement, though many think they got it wrong.

(27:07) **Gary**: That’s interesting. If the artist contributed unique prompts, I’d expect some ownership. This is like the SAG-AFTRA strike—actors fear being replaced by AI-created characters. It’s an evolving issue, and we lack solid answers right now.

(29:14) **Elena**: From a media planning perspective, AI can automate a baseline, but the creative direction still needs human insight. Could the copyright issue create a sort of “artist’s block”?

(31:09) **Gary**: AI amplifies concerns that have always existed in advertising. There are cases where work has raised questions of originality, and AI just heightens these risks. It’s why agencies often use contracts saying they won’t be liable if AI content inadvertently infringes.

(33:17) **David**: Are there clear guidelines on transparency for creators using AI? Is it like sponsored content where disclosure is now expected?

(33:57) **Gary**: Yes, the FTC has been concerned about undisclosed AI usage. Misleading consumers, either by claiming AI involvement or by hiding it, could lead to deception claims. I had a client who wanted to market work as AI-created for novelty, though it wasn’t. I advised against it due to potential deception risks.

(36:42) **Laura**: For a compliance tool we’re developing, how transparent should we be in showing how AI makes decisions? Would this transparency help build trust?

(37

:16) **Gary**: Transparency can build trust, but giving away too much may reveal proprietary information. If your platform can assure clients of AI’s reliability, for example by saying outputs are based on vetted data, it can help ease concerns.

(39:00) **David**: What if a government AI chatbot gives incorrect info that leads to harm? How would liability work?

(40:18) **Gary**: It depends on the duty to users. Government agencies have some protections, though misinformation from an AI chatbot could open liability if it causes harm. For example, a physician using an AI tool to advise patients could face malpractice claims if the advice was faulty.

(41:52) **Thomas**: How effective are watermarks as a solution for AI-generated content?

(42:30) **Gary**: Watermarks may be helpful for transparency, letting consumers know content is AI-generated so they can evaluate it accordingly. California even requires chatbots to disclose they’re not human, which is a similar concept.

(44:03) **Paul**: There’s a new bill against impersonating others using AI. How feasible is it to enforce such a law as AI technology becomes more common?

(45:06) **Gary**: Federal action on AI laws is unlikely in the near term, but I could see states like California leading the way, which might effectively create a national standard.

(45:45) **David**: If I create ad copy and use AI to improve a few sentences, do I have to disclose that?

(46:21) **Gary**: If it’s just a few words, it’s likely considered inspiration. However, the Copyright Office advises disclosing AI-generated portions in registered works.

(47:47) **Thomas**: What about tools like Grammarly? At what point does AI involvement require disclosure?

(48:21) **Gary**: The Copyright Office is currently collecting public input to establish AI guidelines, but clarity is still evolving.

(50:00) **Paul**: Does a prompt itself hold copyright value, as it’s essential to the AI process?

(50:10) **Gary**: Not necessarily. For instance, instructing someone to take a photo doesn’t grant copyright to the director; it still belongs to the creator of the output.