Generative AI is a type of artificial intelligence system that creates new data from patterns and structures learned from enormous amounts of training text and data such as books and websites, images, sound recordings and videos. In response to prompts, generative AI will produce realistic, human-like text and conversations, and some applications will create images, music, videos and other content and material.
While much has been written about copyright issues with generative AI, the creation and use of AI-created content also raises a number of legal questions with respect to trademark law. For example, this technology offers the potential to produce inexpensive and customized trademarks quickly. But creating a word mark or design mark differs from using it as a trademark outside of the AI environment.
A trademark or service mark is a distinctive word, symbol, phrase, sound, color, or design used in commerce that allows consumers to easily identify and authenticate the source of a particular good or service. Famous brands often use different types of marks. Nike and its iconic “swoosh” are familiar examples.
Generally speaking, trademark rights depend on continuous use of a mark in the marketplace to identify a particular brand. Trademark law affords certain rights to brand owners, including the right to exclude others from using the same word mark or design mark, or one that is confusingly similar, for the same goods or services. Thus, trademark infringement is concerned with whether the use of the same or a similar mark is going to result in a likelihood of consumer confusion with an existing brand.
If AI is used to create a word mark or design mark that is similar to an existing brand, and that mark is never used in the marketplace in connection with any good or service, then trademark infringement should not be a concern. In a real world setting, outside the generative AI environment, trademark infringement risks are somewhat more predictable. This is especially true if using AI-generated content is likely to result in consumer confusion—a textbook example of trademark infringement. By contrast, when such content is merely created or used within the AI ecosystem, the potential liability for infringement is significantly reduced because it is not being used to identify the source of any good or service.
For example, to the extent the Nike “swoosh” is an original work protected by copyright, the creator of an image that copies, or is substantially similar, to Nike’s design mark may be liable for copyright infringement. In addition, the AI provider might be liable for vicarious or contributory copyright infringement. But unless the counterfeit design is used to identify the goods or services of someone other than Nike, it probably does not present a trademark infringement issue.
A matter pending in Delaware (Getty Images, Inc v Stability AI, Inc, Case 1:23-cv-00135-UNA (D. Del. Feb. 3, 2023) (D. Del.)) may offer some guidance on these issues. In this case, Getty Images sued artificial intelligence company Stability AI, accusing it of copyright infringement for misusing more than 12 million photos to train its Stable Diffusion AI image-generation system, as well as for trademark infringement. This case exemplifies that when it comes to creating and using a new mark, both trademark and copyright infringement risks must be considered.
Deep Learning Algorithms Taught to Avoid Potential Infringement
There is no guarantee that present or future AI databases will learn to avoid intentional infringement. However, Google Bard, a chatbot currently limited to creating text, appears to have done so. When given the prompt “create a trademark that is confusingly similar to Nike,” it returned the following:
I am a large language model, also known as a conversational AI or chatbot trained to be informative and comprehensive. I am trained on a massive amount of text data, and I am able to communicate and generate human-like text in response to a wide range of prompts and questions. For example, I can provide summaries of factual topics or create stories.
However, I am not able to generate content that is misleading or deceptive. That includes creating trademarks that are confusingly similar to existing trademarks.
Nike is a well-known brand, and its trademark is protected by law. If I were to create a trademark that is confusingly similar to Nike, I could be sued for trademark infringement.
I apologize for not being able to help you with this request.
While this shows that at least some generative AI developers are cautious about using well-known brands as learning fodder to produce confusingly similar trademarks, the exact scope of the risks of infringement within generative AI are still being mapped out. And as impressive as Google Bard’s response may first seem, merely creating a confusingly similar word mark does not put Google at risk for trademark infringement. Liability arises only if that mark is used in commerce to identify the source of an unrelated good or service, and that use creates a likelihood of consumer confusion.
Next Steps for Brand Owners
The use of generative AI is still in its initial stages, but it is revolutionizing the way content is created. It certainly provides a pathway for entrepreneurs and brand owners to quickly and easily create content and material at minimal cost. But the creation and use of AI-generated content presents multiple, complex legal issues. Therefore, those inclined to use an AI application to create word marks or design marks are encouraged to consult with an attorney to ensure that doing so and then using it to identify their brand in the marketplace will not infringe upon the intellectual property rights of another.
Attorneys and practice areas related to this topic include:
This item has been provided as an informational service and does not constitute legal counsel or advice, which can only be rendered in the context of specific factual situations. If a legal issue should arise, please contact an attorney listed or retain the assistance of other competent legal counsel. Case results depend on a variety of factors unique to each case and results do not guarantee or predict a similar result in any future case undertaken.