Advances in technology, and the inevitable litigation that follows, often unfold against a backdrop of long-standing laws—and artificial intelligence (or AI) will not be an exception. For an example of AI-related disputes being litigated under an established legal framework, one only has to look to the application of copyright laws largely in place for over a century to generative AI lawsuits brought by Sarah Silverman and other artists and media companies. See, e.g., Peter Brown, “ChatGPT May Be Threatening Your Copyrights and Privacy,” New York Law Journal, Aug. 7, 2023. While perhaps not as attention grabbing as those copyright cases, trade secret law—developed over decades through common law decisions and state and federal statutes—will likely be one of the key areas of law where stakeholders work to protect, or in some cases protect against, evolving AI technology.

This article examines the core elements of traditional trade secret misappropriation claims and their likely application to AI technology that continues to advance at seemingly warp speeds. Section I takes a practical approach to thinking about the definition of AI. Section II discusses the definition and requirements for establishing an enforceable “trade secret,” and how those requirements are likely to be applied with respect to AI innovations. Section III then discusses how one type of trade secret known as a “combination trade secret” could play a critical role in future AI-related trade secret litigation.
|

An Overview of Artificial Intelligence

Definitions of “artificial intelligence” continue to vary (and be debated), but AI commonly refers to computer systems that are capable of performing tasks that mimic human intelligence. See “The Artificial Intelligence Glossary,” Sept. 30, 2024, https://www.law.com/legaltechnews/2024/09/30/the-artificial-intelligence-glossary/. The term “artificial intelligence” also serves as something of an umbrella label encompassing a variety of more specific technologies, including machine learning and deep learning. Id. Unsurprisingly, AI technology can be inherently valuable, and can also drive practical use cases, including the following examples (recently discussed by the U.S. Patent and Trademark Office):

• Anomaly detection whereby the use of a neural network can identify or detect anomalies (and can be used in various industries like cybersecurity) (PTO Example 47).
• Speech separation in which AI technology analyzes speech signals and can separate desired speech from extraneous or background speech capable (PTO Example 48).
• Fibrosis treatment in which an AI model has been designed to assist in personalizing individual patient medical treatment. (PTO Example 49)

See, https://www.federalregister.gov/documents/2024/07/17/2024-15377/2024-guidance-update-on-patent-subject-matter-eligibility-including-on-artificial-intelligence.

What makes the examples above, and many real world applications of AI challenging, is that the AI will likely be part of complex, multi-component systems of which some of the parts may already be known but where certain other components may be the “secret sauce” to the new, innovative AI feature. In this fast-paced setting, the protection of AI-related innovations presents a complex question: under what conditions will trade secret law be available and what issues are likely to get litigated?
|

What’s the Definition of a Trade Secret, and Can AI-related Innovations Be Trade Secrets? Yes.

There is a well-established body of law in the United States addressing misappropriation of trade secrets, originally built through common law decisions (which, for example, New York state courts continue to follow), state statutes (with the vast majority of states having adopted the Uniform Trade Secrets Act) and finally at the federal level by way of the Defend Trade Secrets Act (DTSA). See, e.g., Brian Yeh, “Protection of Trade Secrets: Overview of Current Law and Legislation,” Congressional Research Service, April 22, 2016. While at a more granular level there can be differences among the common law, state law, and federal law sources, there is a common set of elements involved in any misappropriation of trade secret claim: subject matter that qualifies for trade secret protection; misappropriation of that trade secret by another party; and damages or the threat of damage to the party claiming the trade secret. This article focuses almost exclusively on the first element and will analyze it by looking primarily to the provisions of the DTSA.

The good news for those seeking trade secret protection is that the statutory definition of a “trade secret,” in the first instance, is broad. For example, at the federal level, “trade secret” is: “all forms and types of financial, business, scientific, technical, economic, or engineering information, including patterns, plans, compilations, program devices, formulas, designs, prototypes, methods, techniques, processes, procedures, programs, or codes, whether tangible or intangible, and whether or how stored, compiled, or memorialized physically, electronically, graphically, photographically, or in writing ... .” 18 U.S.C. s. 1839(3). Given the breadth of this definition, there has not been much serious debate as to whether many AI-related innovations fit within within the statutory definition of a trade secret. See Erik Weibust and Dean Pelleteir, “Protecting AI-Generated Inventions as Trade Secrets Requires Protecting the Generative AI as Well,” IP Watchdog, July 24, 2022, (providing an overview of patent versus trade secret law protections, and discussion of arguments for trade secret protection with AI innovations).

However, even if the AI innovation fits within the broad statutory trade secret list above, there are three more requirements to meet before one can enforce an AI-related trade secret:

Independent Economic Value. One claiming trade secret protection has to establish that “the information derives independent economic value, actual or potential, from not being generally known to, and not being readily ascertainable through proper means by, another person who can obtain economic value from the disclosure of use of the information.” In short, the trade secret proponent must establish the information has value because it is unknown to others. Establishing the economic value of trade secret information gets practical quickly, and courts look to factors ranging from the extent to which a plaintiff can show economic value from keeping the information secret to the money that a competitor could save if using the trade secret. See, e,g., Synopsis v. Risk Based Security, 70 F.4th 759, 771-72 (4th Cir. 2023).

In the context of AI-innovations, establish independent economic value should be the most direct application of traditional trade secret case law. For example, even if certain aspects of the AI model used in conjunction with the innovator’s efforts are not transparent, that innovator needs to be able to isolate and value the economic effort that results from that trade secret. Within the context of a trade secrets trial, this will likely be established by way of an economic expert; however, steps taken early in the litigation to identify or segregate that revenue associated with the trade secret will assist expert witness efforts (and also start a damages discussion).

Reasonable Steps to Keep Information Secret. A trade secret plaintiff must also establish that “the owner has taken reasonable measures to keep such information secret.” 18 U.S.C. § 1839(3)(B). Courts and commentators have long identified a litany of factors to consider when determining if reasonable steps have been taken to maintain secrecy. See, e.g., Amy Bergeron, “For Good Reason: ‘Reasonable Measures’ in Recent Trade Secret Law,” June 25, 2021. Most commonly cited among these are the use of confidentiality or nondisclosure agreements that define and protect confidential information/trade secrets, employee policies (e.g., handbooks) showing steps taken by the company to maintain trade secrets, and steps like password protection and access restrictions to protect trade secrets. Applied to AI-related innovations, again the theme of isolating publicly available information and collecting information to show how the innovator has protected the trade secrets will be critical. Innovators will need, in particular, to keep in mind what steps make sense to protect AI-related technology when the underlying technology itself is evolving so quickly.

Identifying Trade Secrets with Sufficient Particularity. The third requirement for protecting a trade secret is already being litigated in the context of AI innovations: whether a plaintiff has identified the alleged trade secrets with sufficient specificity to inform the defendant of what information is at issue. For example, in Yasmine v. Toolbox for HR Spolka z Organiczona Odpowiedzialnoscia Spolka Komandytowa, 2023 WL 6259412, *6 (D. Ariz. Aug. 8, 2023), plaintiff claimed it had developed trade secrets “pertaining to its development of various processes that automate the work of recruiters with enhancements in artificial intelligence, which include patterns to make enhanced prognostications … .” The court disagreed, holding that the assertion only related to “early groundwork” and nowhere identified the processes that could lead to “enhanced prognostications.” Id. Other federal district courts have similarly rejected attempts by plaintiff to satisfy the sufficient particularity requirement with AI-related innovations on the grounds that the assertions are simply too general. See Vortexa v. Cacioppo, 2024 WL 2979313, *10 (S.D.N.Y., June 12, 2024) (rejecting plaintiff’s argument that a document containing information on plaintiffs “methodology and AI/ML models” established a protected trade secret on a preliminary injunction motion where a competing affidavit showed defendant had “already incorporated its own artificial-intelligence and machine-learning models into its own software”); T2 Modus v. Williams-Arowolo, 2023 WL 6221429, *5 (E.D. Tex., Sept. 25, 2023) (rejecting plaintiff’s use of “conclusory terms” such as “artificial intelligence,” “machine learning,” or “proprietary software” without “additional specific information”). These cases all strongly suggest that it is not enough to say a trade secret has something to do with AI and expect that courts will require no further specificity.
|

An Additional Trade Secret Possibility for AI Innovations: The Combination Trade Secret

What happens if an AI innovation involves multiple components or parts, some of which are public and others that are secrets that give the owner an economic advantage in the marketplace? Imagine an AI-powered customer service system that uses a natural language processing model, but then is integrated with proprietary data sets and unique analysis algorithms. Or consider a predictive maintenance tool for manufacturing that could combine relatively standard or well-known predictive algorithms with real-time proprietary data and processing techniques. Finally, consider an AI-driven personalized marketing tool that integrates an open-source recommendation engine with custom machine learning models to deliver targeted marketing campaigns. In each of these cases, while certain individual components may be public, the specific way they are combined arguably creates unique economic value and competitive advantage.

Well-established trade secret law may have a creative solution for protecting such AI innovations: the combination trade secret. The Second Circuit has long recognized the validity of combination trade secrets: “a trade secret can exist in a combination of characteristics and components, each of which, by itself, is in the public domain, but the unified process, design and operation of which, in unique combination, affords a competitive advantage and is a protectable secret.” See Imperial Chemical Industries v. National Distillers & Chemical, 342 F.2d 737, 742 (2d Cir. 1965). A combination trade secret may protect a specific set of elements—such as AI-related algorithms, datasets, and processing methods—that individually might not be a protectable secret but, when used together in a unique way, create a competitive economic advantage. This could give the AI innovator a possible path to protect the technology as a trade secret by isolating those components that might not be confidential, and focusing on steps to protect the confidential components and maximize the economic value derived from that combination.

Asserting a combination trade secret impacts the three traditional requirements for trade secret protection discussed above, particularly in the context of AI. First, the independent economic value must arise from the combination, in which the combination provides a significant competitive advantage that would not be obvious to competitors. Second, reasonable efforts to maintain secrecy focus on protecting the combination as a whole, rather than each individual element, through measures like restricted access, confidentiality agreements, and technology safeguards. Third, in the context of the combination trade secret, the proponent will need to identify the combination with enough specificity for the accused to understand the nature of that trade secret: “A plaintiff asserting a combination trade secret over highly complex technical information cannot merely offer lists of broad technical concepts identifying categories of information without showing which information contained within those categories constituted a trade secret.” See Caudill Seed & Warehouse Co. v. Jarrow Formulas, 53 F.4th 368, 381 (6th Cir. 2022). Each of these three requirements when applied to AI innovations can create significant planning challenges that need to be understood heading into litigation.
|

Conclusion

As AI innovations continue to expand in the marketplace, innovators must prioritize strategies to protect their AI-related intellectual property. For those relying on trade secret law, it is critical to understand the core elements of trade secret litigation and how they apply to this rapidly advancing technology. While each case will depend on specific facts and circumstances, taking steps now to clearly identify trade secrets, their economic value, and the measures implemented to protect them will make a real difference if and when litigation arises.

Rich Harper is the head of Baker Botts’ New York office and co-head of the firm’s artificial intelligence practice group.