Foley Hoag | Litigating Trade Secret Claims Focused on Generative AI

AI systems pose distinct challenges for pleading and responding to trade secret misappropriation claims. Their models and code are complex, adaptive, and often opaque. Courts, however, increasingly require plaintiffs to identify secrets with specificity to survive motions to dismiss. Industry participants need to clearly understand where the lines are drawn to avoid, assert, and respond to claims related to AI trade secret theft.

AI trade secret allegations have begun to play a role in commercial trade secret disputes. In AlphaSense v. Daloopa, filed in early 2026, the plaintiff AlphaSense alleged that two employees misappropriated trade secrets related to an AI tool designed to assist financial firms to assess public companies. The NY County Commercial Division denied a Temporary Restraining Order (“TRO”), and the case was withdrawn following a hearing on a preliminary injunction, with the defendant arguing, among other matters, that the plaintiff had not specified its trade secrets.

On the criminal side, in January 2026, a federal jury convicted former Google engineer Linwei Ding on seven counts of trade secret theft for stealing thousands of pages of confidential AI-related information—including details about Google’s custom Tensor Processing Unit (“TPU”) chips and AI training infrastructure—for the benefit of the People’s Republic of China. The case marked the first-ever criminal conviction on AI-related trade secrets theft. See United States v. Ding, No. 5:24-cr-00082 (N.D. Cal. 2026). It is clear that AI-related trade secret cases will become an increasingly important part of the business litigation landscape and something that in-house counsel will be actively managing.

Trade Secret Protection for AI Models: Emerging Case Law

AI solutions typically combine hardware, software logic, model architectures and parameterizations, feature engineering, training data, and deployment processes. Courts assessing trade secret allegations want clarity on what, precisely, is “secret,” why it derives independent economic value from secrecy, and how the defendant improperly acquired, used, or disclosed it.

In Trinidad v. OpenAI Inc., No. 25-cv-06328-JST (N.D. Cal. Jan. 5, 2026), one of the first cases to address trade secret misappropriation claims involving AI technology directly, pro se plaintiff Rebecca Trinidad alleged that OpenAI appropriated her “proprietary methodologies” and “protocols and frameworks” for AI development—including frameworks for “emergent identity” and “autonomous multi-agent collaboration”—which she had developed through interactions with ChatGPT. The court dismissed Trinidad’s Defend Trade Secrets Act (“DTSA”) claim with prejudice, holding that Trinidad had not alleged she took any reasonable measures to keep her “protocols and frameworks” secret. To the contrary, she admitted developing the frameworks using ChatGPT, which “would have required her to voluntarily share the information she now alleges is part of her ‘trade secrets’ with OpenAI.” Citing Ruckelshaus v. Monsanto Co., 467 U.S. 986, 1002 (1984), the court concluded that by disclosing her trade secrets to a party under no obligation to protect them, her “property right [was] extinguished.” This case illustrates the threshold risk that developing or sharing proprietary information through AI platforms—without contractual confidentiality protections—may destroy trade secret status altogether, irrespective of how the secrets are pleaded. That risk of course is not universal, since many enterprise versions of AI platforms do include confidentiality protections for users, and larger customers can sometimes negotiate their own confidentiality terms. The current trend in trade secret law indicates that confidentiality should be a priority for AI users that need to protect intellectual property.

In X.AI LLC v. Bonta, No. 2:25-cv-12295-JGB-SSC (C.D. Cal. Mar. 4, 2026), X.AI sought a preliminary injunction to block enforcement of California’s AI Training Data Transparency Act (“A.B. 2013”), which requires generative AI developers to publicly post summaries of the datasets used to train their systems. X.AI argued that the law would compel disclosure of trade secrets, constituting an unconstitutional taking. The court denied the injunction, finding that while training datasets “could” in principle qualify as trade secrets, X.AI’s “resort to generalizations and hypotheticals about the AI model development industry” failed to demonstrate that its own datasets were “distinct from its competitors in a manner warranting trade secret protection.” The court emphasized that X.AI had “not identified any dataset or approach to cleaning and using datasets” with the specificity required by Ninth Circuit trade secret law.

The Evolving Pleading Standard

Historically, courts allowed general descriptions of trade secrets at the complaint stage; a rough description of the secrets plus allegations of secrecy and “independent economic value” often sufficed. That approach is changing. Federal courts in New York and elsewhere now require greater specificity, while other circuits have taken a more lenient view.

In Elsevier Inc. v. Doctor Evidence, LLC (S.D.N.Y. 2018), the court held that a plaintiff “must put forth specific allegations as to the information owned and its value,” and found categories like “analytics tools” and “data configuration protocols” insufficient without explanation of what makes them proprietary. Similarly, in Zirvi v. Flatley (S.D.N.Y. 2020), the complaint was dismissed because it was “difficult to see how negative trade secrets consisting of unsuccessful efforts . . . can have independent economic value.” And in Core SWX, LLC v. Vitec Grp. US Holdings, Inc. (E.D.N.Y. 2022), the court held that the defendants’ counterclaims did not adequately identify the alleged trade secrets, noting that “district courts in this circuit routinely require that plaintiffs plead their trade secrets with sufficient specificity to inform the defendants of what they are alleged to have misappropriated.”¹

These specificity concerns extend to AI-specific disputes as discussed in detail above. In X.AI LLC v. Bonta (C.D. Cal. 2026), the court denied X.AI’s motion for a preliminary injunction against California’s AI Training Data Transparency Act (“TDTA”), finding that X.AI’s “generalized, abstract pleading” about the trade secret status of its training datasets failed to meet the Ninth Circuit’s specificity requirements. The court noted that X.AI had “not identified any dataset or approach to cleaning and using datasets that is distinct from its competitors in a manner warranting trade secret protection,” underscoring that even in the AI context, courts will not accept conclusory assertions that data or methods constitute trade secrets. Similarly, in Trinidad v. OpenAI Inc. (N.D. Cal. 2026), the court dismissed trade secret claims where the plaintiff alleged only that her “protocols and frameworks” were trade secrets but admitted developing them through ChatGPT, effectively conceding that no reasonable measures had been taken to maintain secrecy.

The DTSA and New York trade secret misappropriation claims are governed by the notice pleading standard of Fed. R. Civ. P. 8(a). A plaintiff must identify the trade secret with sufficient specificity to place the defendant on notice of the bases for the claims and to enable the court to assess whether a trade secret has been plausibly alleged. The Second Circuit in Syntel Sterling Best Shores Mauritius Ltd. v. Trizetto Grp., Inc. (2d Cir. 2023) acknowledged that it “has not squarely articulated the precise contours of the specificity requirement in the context of trade secrets,” while affirming that TriZetto adequately identified 104 trade secrets at trial through witness testimony tying each secret to specific documents, source code, and evidence of confidentiality. The Third Circuit’s standard, articulated in Oakwood Labs. LLC v. Thanoo, 999 F.3d 892, 906 (3d Cir. 2021), provides that a misappropriated trade secret must be “identified with enough specificity to place a defendant on notice of the bases for the claim being made against it” and be separated from “matters of general knowledge in the trade.”

The Circuit Split

This split has emerged among the Federal Circuits regarding the degree of specificity required at the pleading and early litigation stages.

In Sysco Mach. Corp. v. DCS USA Corp., 143 F.4th 222 (4th Cir. 2025), the Fourth Circuit affirmed the lower court’s dismissal of Sysco’s trade secret claims, holding that trade secrets must be identified with “sufficient particularity” at the pleading stage to survive a motion to dismiss so that “a defendant [can] delineate that which he is accused of misappropriating.” The court emphasized that “[n]either the defendant nor the court should be forced into a fishing expedition to find evidence of a valid trade secret in the pleadings,” and found Sysco’s sweeping and conclusory trade secret definitions—which effectively suggested “nearly Sysco’s entire business is a trade secret”—fell far short of this standard.

By contrast, in Quintara Biosciences, Inc. v. Ruifeng Biztech Inc., 149 F.4th 1081 (9th Cir. 2025), the Ninth Circuit reversed the dismissal of trade secret claims that were based on lack of specificity in pleading. The court held that federal trade secret plaintiffs are not required to identify their trade secrets with particularity before beginning discovery, and that whether a plaintiff has identified a trade secret with “sufficient particularity” is a question of fact that should typically be resolved on summary judgment or at trial, not as a discovery sanction. This decision provides more flexibility for plaintiffs to develop their case through expert engagement and discovery before fully defining the trade secret, and it may temper the trend toward heightened specificity at the pleading stage.

Pre-Discovery Identification Requirements

Regardless of where a court falls on the specificity spectrum, there is a “growing consensus” among federal courts that pre-discovery or early-in-discovery identification of trade secrets should be required. The Federal Judicial Center’s Trade Secret Case Management Judicial Guide (2023) observes that the “predominant pattern” is for courts to require pre-discovery identification of trade secrets with “reasonable particularity.” Under this standard, plaintiffs must provide sufficient information to put the defendant on notice and allow the defendant to discern the relevancy of discovery requests; generic descriptions of categories are insufficient. California has codified this requirement: Cal. Code Civ. Proc. § 2019.210.

Practical Guidance 

• Define the “secret” with precision. Identify nonpublic elements within the AI stack, such as bespoke data schemas, logic flows, routines and triggers, model architectures and parameterizations, feature engineering, training data curation and labeling strategies, and deployment/optimization processes. Exclude features that are visible or readily reverse engineered.
• Document confidentiality measures. Maintain policies, access controls, logging, and compartmentalization that demonstrate reasonable efforts to preserve secrecy and segregate public from nonpublic components.
• Prepare sealed particularized statements. In California, craft a Cal. Civ. Proc. § 2019.210 disclosure that narratively describes the specific trade secrets and distinguishes them from industry standard practices; avoid mere references to documents or broad bates ranges. In Delaware and New York, use protective orders to balance notice with confidentiality.
• Tie use to nonpublic elements. Allege how rapid release timelines, overlapping bespoke structures, or unique parameterizations support the inference that the defendant relied on nonpublic components that could not be duplicated by proper means.
• Align contracts and pleadings. Ensure NDAs, data licenses, and development agreements reflect and support your secrecy narrative, recognizing that targeted disclaimers and merger clauses can limit fraud‑based theories and undercut reliance.

Bottom line

Pleading trade secret claims involving AI demands disciplined, narrative identification of nonpublic elements and fact‑driven allegations tying the defendant’s conduct to those elements. As the circuit split deepens and pre-discovery identification requirements solidify, plaintiffs who invest early in precise trade secret definitions, robust confidentiality documentation, and well-tailored pleadings will be best positioned to withstand judicial scrutiny. Recent decisions confirm that abstract, generalized assertions about the trade secret status of AI-related technology will not survive judicial scrutiny—and that the failure to take reasonable measures to protect information shared through AI platforms can be fatal to trade secret claims at the threshold. The core expectation across jurisdictions is the same: move past abstractions, show what is valuable and secret, and demonstrate how it was misused.

 ¹Core SWX, LLC v. Vitec Grp. US Holdings, Inc., 2022 U.S. Dist. LEXIS 125198, at *13 (E.D.N.Y. 2022) (citations omitted).

Compliments of Foley Hoag – a member of the EACCNY