Trade Secrets and Artificial Intelligence Difficulties and Solutions
July 15, 2026 New York Law Journal
By Jonathan Bick
Jonathan Bick is counsel at Brach Eichler in Roseland, and chairman of the firm’s patent, intellectual property, and information technology group. He is also an adjunct professor at Pace and Rutgers Law School.
Artificial intelligence (AI) is remaking the intellectual property environment. Just as the output from generative AI has altered patent and copyright intellectual property law, the input to and use of machine learning AI and generative AI have transformed trade secret law. Trade secret content disclosed in machine learning training and in AI prompts will likely compromise trade secret enforceability. Legal, business and technology solutions are available to ameliorate AI trade secret difficulties.
AI comprises a variety of forms (e.g., machine learning, deep learning, natural language processing). Each needs input data and generates output data. For instance, machine learning AI involves training data input to “teach” the AI to recognize patterns in a dataset, as well as testing data input to assess the model’s accuracy. Generative AI requires inputs called prompts.
United States trade secret law is often understood to limit trade secret protection to information which is kept secret. Since 1939 for example, the Restatement (First) of Torts focused on secrecy being the pivotal element, clarifying that “[t]he subject matter of a trade secret must be secret” (see Restatement of Torts section 757, comment b (1939)).
Trade secrets are shielded by numerous statutes such as the Economic Espionage Act of 1996 (18 U.S.C. §§ 1831-1839 (2022)) which enables criminal prosecution and civil claims by the U.S. attorney general for stolen trade secrets, as well as a noteworthy amount of case law such as Ruckleshaus v. Monsanto Co., 467 U.S. 986 (1984) where the Supreme Court found that trade secrets are property rights if maintained as property.
Trade secrets, unlike patents, lose their legal protection due to public disclosure. Trade secret protection lasts so long as the information is kept secret.
Disclosure of a trade secret occurs when information qualifying as a trade secret is revealed to a party without the consent of the owner. Such acquisition of a trade secret may be due to sharing information with unauthorized individuals intentionally or accidentally. For example, by making trade secret content accessible to an entity not bound by confidentiality consent.
Increasingly, as the use of AI tools becomes widely available, employees and researchers use AI tools to develop or refine proprietary information. Most often, both AI inputs for AI machine learning and generative AI prompts and all AI outputs are normally stored by the AI software to be used to train or fine tune the AI model itself. From a technical perspective this activity threatens to reveal trade secret information, therefore undermining trade secret claims.
AI systems can disclose trade secrets in several ways. For example, if an AI system is trained on data that includes trade secrets, it may generate outputs that reveal this confidential information to users or third parties. Alternatively, AI models often require large datasets for training. If trade secret information is included in the training data and the model is not properly secured, there is a risk that the model could be reverse-engineered or probed to extract the underlying confidential information.
Unauthorized use and/or sharing may also result in an AI transaction that yields trade secret legal difficulties. For instance, when access controls are insufficient, individuals without trade secret disclosure authorization may use AI systems to directly or indirectly transfer trade secret content. This can occur if the AI system is integrated with sensitive databases and those databases are not properly monitored. Similarly, when AI systems are developed or operated by third-party vendors, there is a risk that trade secrets could be disclosed through vendor access, insufficient contractual protections, or inadvertent integration with external platforms.
AI may undermine trade secrets in a completely different way, namely when AI systems make trade secrets easier to challenge. AI tools are increasingly capable of inferring hidden information from public sources, by being trained on redacted documents and being tasked to guess what lies under the redaction. Thus, third parties regularly input patents, publications, product data, and even redacted court documents into AI tools and prompt the AI to “fill in the gaps.”
Reverse engineering is generally a valid, lawful method for learning trade secrets in the United States and many other jurisdictions, provided the product or information was acquired honestly. Under both the Uniform Trade Secrets Act (UTSA) and the Defend Trade Secrets Act (DTSA), analyzing a publicly available product to understand how it works is not considered “misappropriation.” Trade secret case defendants use this type of AI-enabled reverse-engineering to argue that a claimed trade secret was easily discoverable and hence not subject to trade secret protection.
On the other hand, trade secret holders use AI to initiate new discovery battles in litigation. Parties now seek extensive discovery on each other’s AI activity – policies, tools, prompts, outputs, logs, and historical usage – including employees’ and contractors’ usage at home, on their personal devices to demonstrate that intellectual property claiming trade secret protection, was disclosed.
AI will also likely reshape employee mobility and “memory” trade secret disputes. Employee departures regularly generate litigation over what information an employee allegedly misappropriated as part of their departure. AI workplace tools have obscured the line between an employee’s personal know-how and company-owned information. The more employees rely on AI tools to sketch out code, devise systems, or produce business strategies, the more questions could emerge about whether the AI outputs were effectively stored, logged, or traceable long after the employee leaves. Litigation will arise over whether an employee’s use of AI expanded, preserved, or unfairly transferred knowledge that would otherwise exist only in human memory – and over who owns the human memory that remembers how to prompt the AI.
AI has also created a category of “derived” or “model-dependent” trade secrets. Historically, trade secrets have involved human-created information: formulas, processes, customer data, strategies. Courts now must determine if AI generated knowledge (and the such “derived” or “model-dependent” content) qualify as a protectable trade secrets. Disputes have already arisen as to whether the weights, training data selections, embeddings, or emergent insights of proprietary models constitute trade secrets, and whether defendants can be liable for misappropriation when they never accessed the underlying inputs directly. Companies and practitioners can expect this new frontier to test the limits of traditional trade secret doctrines.
Legal, business and technology solutions should be considered to eliminate or ameliorate AI trade secret difficulties. Said difficulties may arise due to internal or external AI tool use.
For example, an external AI tool use may result in an AI trade secret difficulty when a third party uses AI independently and discovers or reverse-engineers a trade secret and uses it for their own purposes. This often occurs for software algorithms.
One potential legal work around for trade secrets which might be discovered by AI is memorizing the trade secret in a copyright registration. This is particularly useful for software related trade secrets which are incorporated into algorithms.
17 U.S. Code § 102 discloses what content is subject to copyright protection. Circular 61 Copyright Registration of Computer Programs (https://www.copyright.gov/circs/circ61.pdf last visited 5/6/2026) states that copyright protection for a computer program extends to all of the copyrightable expression embodied in the program, thus, an algorithm could be copyrightable.
Another potential legal work-around for trade secrets which might be discovered by AI is patenting the trade secrets. Method patents, also called process patents, are regularly allowed by the United States Patent Office to specifically protect innovative processes and methods,
Technical solutions are available to protect trade secrets from AI related difficulties. For example, entities may store trade secrets in a virtual vault. Such a vault typically requires a two- factor authentication access code, access and use monitoring, and isolation from AI systems.
Technological solutions are also available to ameliorate the inadvertent disclosure of trade secrets due to AI exposure. For example, to protect and maintain trade secrets, entities may use special digital encryption system that digitally fingerprint/watermark confidential information within internal storage. Such encryption may be adapted to prevent the content from being transferred to either AI machine learning or AI generative systems. Such encryption is usually adapted to resist alteration, and can be used as evidence in court that confidential information existed at a specific time.
Business solutions are also available to ameliorate the inadvertent disclosure of trade secrets due to AI exposure, such as insurance. Traditional insurance policies, such as property or liability insurance, generally do not cover intangible assets like trade secrets.
However, the insurance industry offers products that can mitigate the risks associated with the loss of trade secrets due to AI use. These specialized policies are often categorized under broader intellectual property insurance or cyber liability insurance. These policies can provide coverage for losses resulting from the theft, misappropriation, or accidental disclosure of trade secrets.