What brand owners need to know to protect their trademark rights in the era of generative AI

Since the launch of OpenAI's ChatGPT in November 2022, businesses and consumers alike have been captivated by the promise of generative artificial intelligence (AI). Platforms like ChatGPT and its competitors offer users the ability to generate text, software code, images, and even video from simple prompts.

As with many new technologies, the potential benefits of generative AI also bring legal risks for companies and brands as they create and use generative AI models. While much has been written about the potential copyright implications of these generative AI models, many of the most prominent cases also include trademark infringement allegations, and the use of generative AI by businesses implicates a number of trademark law concerns. Risks include the potential weakening of trademark rights when a brand's trademarks are used by the generative AI platforms. But the technology also offers potential benefits: not only can generative AI create useful content for brands, emerging AI-based tools can also be used to more efficiently search for potential instances of infringement. Brands must be mindful to acknowledge and weigh both the benefits and risks.

THE LEGAL LANDSCAPE: BRAND OWNERS' ENFORCEMENT OF TRADEMARK RIGHTS AGAINST GENERATIVE AI PLATFORMS

In early 2023, stock photo provider Getty Images sued Stability AI over its Stable Diffusion image generation platform, arguing, among other things, that some of the images created by the generative AI platform included distorted (but legible) versions of the Getty Images trademark in the same place and format as the real Getty Images watermark appears on authentic photos. Getty Images alleges that these outputs are not only infringing Getty Images' trademarks, but also diluting the reputation of Getty Images for high-quality images, particularly because the Stable Diffusion output images many times are unrealistic or distorted, unlike authentic Getty Images photos. In simpler terms, the gist of the complaint is that the generative AI platform is creating distorted images containing versions of the Getty Images watermark, potentially confusing consumers into believing that the low-quality and sometimes strange images come from Getty.

In December 2023, the New York Times brought a similar suit against Microsoft and OpenAI, alleging that their generative AI platform is creating false news articles and false product reviews attributed to the Times and its Wire cutter product review blog (including, for example, one article that linked drinking orange juice to the onset of leukemia). The Times claims that this false information, linked to its name, is harming its reputation as a source of reliable information, and thereby diluting its trademarks.

Both trademark challenges are based on the argument that AI models' outputs are providing users with text and images that violate the plaintiffs' trademark rights. Unlike the copyright claims that have gotten much attention through the successful motions to dismiss waged by the defendants, the trademark claims have not generally been challenged, indicating the AI platforms recognize the claims are at least plausible.

Plaintiffs' trademark infringement and dilution claims center on the idea that the text and images the generative AI platforms are creating use the plaintiffs' trademarks in confusing and dilutive ways. Unlike the copyright claims, the plaintiffs' trademark claims do not relate to AI models' inputs--the training data used to develop the AI model--because no consumer confusion or dilution arises from the internal process of training that consumers using AI models never see. However, each plaintiff claims that the low quality or inaccuracy of the AI outputs, which are being presented alongside the brands' famous trademarks, is somehow tarnishing their brands.

Notably, when Microsoft and OpenAI moved to dismiss the New York Times' lawsuit, they focused on trying to dismiss the copyright claims, and did not attack the trademark dilution claim. This will be an important issue to watch as these cases proceed. While AI platform defendants have had some initial success narrowing copyright claims to those based on the AI model inputs, the viability of the trademark claims suggests that model outputs will continue to face legal challenges. It, of course, remains to be seen whether consumers will truly be confused, or brand value truly diluted, by these generative AI outputs, as no final decision has yet to be reached. And, as with many trademark claims, any success is likely to be case and fact-specific.

THE FUTURE OF GENERATIVE AI: TRADEMARK CONSIDERATIONS FOR BRAND OWNERS

Generative AI could fundamentally change how consumers encounter and learn about brands online. Many in the industry have speculated that generative AI tools will soon replace traditional search engines as the principal way consumers gather information and shop online. As the New York Times case indicates, AI outputs that include trademarks--like attributing news to the New York Times or a product review to the Wirecutter blog--can lead to concerns for brand owners who are not able to control how their brands are used in AI outputs. Unlike traditional search engines, which simply direct users to the brand's website and content, generative AI outputs are typically contained within the platform and created by the AI model.

Brands in the consumer products and fashion spaces should consider how generative AI platforms can be used to support consumer engagement. Generative AI could open entirely new paths for consumers to discover and interact with brands in novel ways. However, brand owners may not able to control how their trademarks (e.g., brand names or logos) or trade dress (the distinctive and protectable aspects of brands' product design) appear in many AI image or video generators. The current versions of most generative AI models return error messages when directly prompted to create registered trademark content by referring to existing brands, but those protections alone may not be an adequate stopgap in all circumstances. As the Getty Images and New York Times cases demonstrate, the more common uses of generative AI tools to search for information or create images--without direct prompts to create trademarked content--can still result in content that includes trademarks. Consumers who see strange or made-up AI-generated products associated with a brand may or may not be able to tell whether those are real, authorized products. In addition, AI-based product search engines may have unintended biases from their training data, leading them to artificially favor one brand over another or produce inaccurate product reviews.

Despite the risks that generative AI platforms can pose for brand owners, an emerging wave of AI tools also promises brand owners new abilities to strengthen and police their marks online. Brand owners can make use of AI tools to create marketing content, including generating captivating images and videos using generative AI, with minimal cost and effort. The speed with which AI can create new material offers brands opportunities to engage quickly with current trends and consumer behaviors, which is important with today's fast-moving social media. However, it will be especially important for brands to review and clear AI-generated content before it is used, especially as part of a new mark.

As consumers engage more and more directly with generative AI through chatbots like ChatGPT, opportunities will arise for brands to engage directly with consumers who are seeking information through these AI programs. Brands may want to consider working directly with AI platforms to ensure that models return accurate and appropriate content, and avoid the problems with AI outputs that have led to the legal challenges discussed above.

In addition, new AI-based tools that can search for content similar to existing brands may offer brand owners an increased ability to detect infringement. Although these tools remain imperfect, in some circumstances they may be an improvement over historical options. Brand owners will need to carefully evaluate the tools and their outputs before bringing trademark challenges, as AI models can be prone to error. AI "hallucinations"--output that is simply wrong and made-up by the AI--require careful attention before using the output of an AI tool as the basis for a claim.

WHAT THE FUTURE HOLDS

As trademark cases against AI platforms work through the courts and AI platforms continue to refine and improve their models' outputs, brand owners and trademark lawyers should carefully monitor developments in the months ahead. Generative AI models are improving at a rapid pace and it is clear that AI is here to stay. Capabilities that were merely thought experiments just months ago--like an AI model able to create high-definition video from a simple text prompt--are now a reality.

As these models continue to develop, brand owners will need to carefully monitor whether and how their trademarks can be generated by AI platforms and take steps as necessary to protect how consumers will encounter their brands in generative AI outputs. Brands will also want to take care to monitor how their licensees are using AI-generated content, and should consider revising their license agreements going forward to add appropriate controls around how licensees make use of generative AI. Being attuned to the potential risks of generative AI will put brands in the strongest position to take advantage of the technology's promise and benefits.

Megan K. Bannigan is a partner and member of the Litigation and Intellectual Property & Media Groups, and Christopher S. Ford is a counsel in the Litigation Department and a member of the Intellectual Property Litigation Group and Data Strategy at Debevoise & Plimpton LLP.*