Trademark and generative AI tools
We expect internal marketing departments to increasingly rely on generative AI to prepare creative content, which will yield content that could be protected by trademark law in addition to copyright law.
For example, a generative AI application might be asked to produce a slate of potential new product names, a fresh look for a webpage, a new slogan for an ad campaign, or a short audio signature or jingle to be used when consumers interact with a new product or game. It’s worth remembering that any of these might be protected by trademark law because they could serve as a source indicator for consumers. Trademarks aren’t just the company name and product name; they are also slogans, sound signatures (think, the MGM lion’s roar), packaging designs, and more. When AI is used to generate these signatures, trademark clearance will be even more critical.
Where before, your internal marketing team might intuitively recognize a slogan or sound as already trademarked and steer clear of such arrangements, a trademark generated by AI might be just different enough not to set off any alarm bells during human review. Models trained on trademarked content, however, could generate outputs that infringe existing trademark rights. Trademark clearance, which is already our recommended approach for all new brand indicia, will be especially critical for AI-generated or AI-assisted content. A robust clearance process will provide reassurance that whatever the output of an AI tool looks, reads, or sounds like, that output will be compared back to the trademark register to identify possible conflicting marks before they become a problem in the marketplace. Trademark clearance provides a risk assessment of using the newly generated source indicator so you can move your brand forward with a better understanding of the legal risks.
We also want to remind our clients that trademark issues can come up inside copyrightable pieces of entertainment content. Should this happen to you, we encourage you to reach out to us to evaluate your use and ensure it falls under the category of fair use. It’s worth remembering the major Ninth Circuit decision in ESS Entertainment, where the Grand Theft Auto video game depicted a satirized version of the Play Pen club, and the club sued the game maker for trademark infringement.1
Issues like those in ESS Entertainment are more likely to come up in the context of AI-generated or AI-assisted art, where each element of a video game, movie, or commercial might not get the thoughtful treatment it would otherwise receive if a human were responsible for adding every aspect of the design.
In ESS Entertainment, the court found the use was fair and therefore not infringing, but we highly recommend an outside evaluation before publishing your content to make sure you aren’t putting your business at risk when using generative AI tools to prepare or inspire it.
Patents and AI
Using generative AI to develop products or inventions for patenting presents both opportunities and risks on an unsettled legal landscape. Some argue that generative AI promises to accelerate the development of inventions that benefit society such as life-saving medicines, and that AI should be recognized as an inventor on patents for such inventions. Others say that because AI is not a human, it cannot be an inventor under the patent statutes of most countries. Still others observe that using generative AI to develop products creates the risk of liability for patent infringement because the data used to train generative AI models may include patents or patented functionality.
On the issue of whether AI can be a named patent inventor, the majority of countries that have considered the issue have found that it cannot. Most recently, the United States Supreme Court refused to consider the issue in denying a petition for certiorari of the decision Thaler v. Vidal, 43 F.4th 1207, 1210 (Fed. Cir. 2022). In that decision, the Federal Circuit – the U.S. appellate court that decides issues of patent law – affirmed a lower court’s ruling upholding the United States Patent Office’s decision to deny petitions to name an AI system called Device for Autonomous Bootstrapping of Unified Sentience (DABUS) as a patent inventor. Based on U.S. Supreme Court precedent and language in the U.S. Patent Act, the Federal Circuit affirmed the holding that an inventor must be a natural person. Id. at 1211. However, the court left open the possibility that AI could contribute to a patented invention, stating that it was not addressing “the question of whether inventions made by human beings with the assistance of AI are eligible for patent protection.” Id. at 1213.
The U.S. Patent and Trademark Office (USPTO) held two listening sessions in April and May 2023 on the current state of AI technologies and related inventorship issues. The USPTO asked for input on 11 questions related to AI and patents, including whether U.S. patent law should be changed so that AI systems are eligible to be listed as an inventor and whether the USPTO should require applicants to provide an explanation of contributions AI systems made to inventions claimed in patent applications. Speakers at the sessions largely agreed that AI cannot be a named patent inventor under the U.S. patent laws as currently written. But they disagreed on whether patent applicants should be required to disclose the contributions of AI to an invention that is the subject of a patent application. Policy efforts in this area are ongoing.
As to the risk of patent infringement claims from using generative AI to develop products, this, too is unclear. The data used to train generative AI systems undoubtedly includes patents and content that describes patented functionality. But tracing the output of generative AI to patents included in data used to train the AI model seems highly unlikely to impossible.
Perhaps the issue most everyone can agree on is that current patent laws are not equipped to deal with AI as an inventor of patented inventions. It remains to be seen whether future legislation will achieve clarity on this issue.
Trade secrets and AI
Given the issues with the patentability of AI output, should inventors turn to trade secret protection? Trade secret protection is often used to safeguard unique intellectual property and can be obtained without application or registration. In the context of AI, trade secret protection could include protecting output, data sets, unique algorithms and machine learning techniques.
Is AI protectable as a trade secret?
The U.S. Uniform Trade Secrets Act defines a trade secret as: “a formula, pattern, compilation, program, device, method, technique, or process, that: (i) derives independent economic value, actual or potential, from not being generally known to, and not being readily ascertainable by proper means by, other persons who can obtain economic value from its disclosure or use, and (ii) is the subject of efforts that are reasonable under the circumstances to maintain its secrecy.”2 Trade secret owners can file suit in a U.S. federal court for damages if their trademarks have been misappropriated under the Defend Trade Secrets Act of 2016.3 In the U.S., it is well established that trade secrets are property rights.4
The EU has issued a Council Directive with similar standards to the U.S. with regard to the definition of what constitutes a trade secret.5 However, the Directive generally does not regard trade secrets as property, and most EU states do not classify trade secrets as property or intellectual property.6
A number of issues need to be considered when applying trade secret protection to AI, including:
- Need for secrecy: Trade secrets, by definition, require maintaining secrecy. However, it can be challenging to maintain the secrecy of AI output or systems, especially in collaborative environments or open-source culture where the sharing of information and techniques is common.
- Reverse engineering: A significant drawback to trade secret protection is that it does not protect against reverse engineering. Independent development will allow competitors to legally reverse-engineer an AI system’s results or the system itself.
- Difficult to enforce: It can be challenging to demonstrate that a trade secret has been stolen or misappropriated. For AI companies, this could require proving that a competitor had direct access to their proprietary information, which is often difficult. Outside of the U.S. and EU, many jurisdictions have weak trade secret laws and/or enforcement practices.
- Employee leakage: In a tech-driven field like AI, where talent is in high demand, employees often move from one company to another. These employees may inadvertently or intentionally carry over knowledge or techniques that could be considered trade secrets, which is a risk for companies seeking to protect their intellectual property in this way.
Trade secret best practices
While trade secret protection for AI may be challenging, much of the industry is using trade secret protection and employing a “zero-trust approach.”7 For example, Google, Facebook and Yahoo!’s algorithms – the “secret sauce,” so to speak, for their AI systems, is trade secret protected. Some of the output from those systems, which is exploited for their own commercial use, is also kept secret.
Trade secret protection begins with traditional trade techniques such as limiting access and requiring employees and independent contractors working with AI to sign confidentiality and work-for-hire agreements. IBM, KDDI Research and the National Institute of Informatics have each introduced methods of watermarking deep learning models to help protect algorithms by identifying the owner of the intellectual property.
Yet while trade secret protection for intellectual property that is used by companies internally may be very useful, overall trade secret protection is imperfect, in particular for AI-generated content. Trade secrets protect against misappropriation and unlawful use. Trade secret law was not intended to be used as an instrument to protect intellectual property that is “let out into the wild.” Trade secret protection will only be effective when access to AI outputs and systems is restricted, which means that it may not be helpful where the desired outcome is the commercial exploitation of AI-generated content.
- ESS Entertainment 2000 Inc. v. Rock Star Videos, Inc., 547 F.3d 1095 (9th Cir. 2008).
- Uniform Trade Secrets Act (1985), Section 1.
- Defend Trade Secrets Act of 2016, Pub, L. 114-153, 130 Stat. 376 (2016).
- Ruckleshaus v. Monsanto Co., 467 U.S. 986, 1003-4 (1984).
- Directive 2016/943 of the European Parliament and of the Council of 8 June 2016 on the Protection of Undisclosed Know-how and Business Information (Trade Secrets) Against their Unlawful Acquisition, Use and Disclosure, OJ, L 157, 1–18;.
- Katarina Foss-Solbrekk, Three routes to protecting AI systems and their algorithms under IP law: The good, the bad and the ugly, Journal of Intellectual Property Law & Practice, Volume 16, Issue 3, March 2021, Page 257, academic.oup.com.
- Stacy Collett, How to Protect Algorithms as Intellectual Property, CSO (July 13, 2020), csoonline.com.
