AI and Trade Secret Protection

Short Answer

AI can create new disclosure paths and make some information easier to reconstruct. Trade secret protection still requires value from secrecy, absence of ready ascertainability through proper means and reasonable protection measures. Comet Technologies USA, Inc. v. XP Power, LLC assigned the Defend Trade Secrets Act (DTSA) plaintiff the ascertainability burden; it did not decide an AI reconstruction issue. In DeWolff v. Pethick, the Fifth Circuit rejected an overbroad database claim by a Dallas-headquartered consulting firm that neither distinguished public from non-public information nor identified the claimed secrets with specificity. A narrow classification improves both proof and practical controls. Whether a particular AI tool can lawfully reconstruct a claimed secret is a factual inquiry, not a conclusion supplied by these decisions.

Which Laws Apply

Texas AI-specific: None.

Generally applicable Texas law: Texas Uniform Trade Secrets Act (TUTSA), Civil Practice and Remedies Code chapter 134A.

Federal: Defend Trade Secrets Act, 18 U.S.C. §§ 1836 and 1839.

Reasonable Measures in an AI Workplace

The measures that count are the ones a business can show: approved tools with confidentiality and no-training terms, controls that block uploads of tagged material, policies employees acknowledged, and training. A business that relies on an employee handbook alone while its staff use public tools freely is exposed. See Employee Use of AI and Confidential Information.

Readily Ascertainable

Information that a competitor could reconstruct through proper means with modest effort is not a trade secret. AI tools lower the effort required to reverse engineer, search and synthesize public information. A defendant may argue that a plaintiff’s “secret” could be generated by asking a model the right questions. Whether that argument succeeds will depend on evidence of what the tool actually produces and at what cost. Courts have not yet adopted it.

The burden question matters. In Comet Technologies USA, Inc. v. XP Power, LLC (9th Cir. July 14, 2026), the court vacated the amended final judgment in its entirety, including compensatory and punitive awards, an injunction and a fee award, because the jury instructions placed on the defendant the burden of proving that the information was readily ascertainable; under the DTSA, the court held, the plaintiff bears it. The court remanded for a new trial; the unchallenged findings, including the verdict for XP on Trade Secrets S and T, remain undisturbed. The decision does not bind Texas courts but is likely to be cited in DTSA cases here.

The Texas definition uses the same test: information that derives independent economic value from not being generally known to, and not being readily ascertainable through proper means by, another person who can obtain economic value from its disclosure or use (Civil Practice and Remedies Code § 134A.002(6)(B)). The Fifth Circuit has applied that test alongside the DTSA definition in a preliminary-injunction setting, affirming the denial of an injunction where customer information in a closed industry was not shown to be a trade secret (CAE Integrated, L.L.C. v. Moov Technologies, Inc., No. 22-50034 (5th Cir. Aug. 9, 2022)). That decision did not address Comet’s jury-instruction question.

Identify Narrowly

In DeWolff, Boberg & Associates v. Pethick, 133 F.4th 448 (5th Cir. 2025), the Fifth Circuit affirmed summary judgment against a management consulting firm headquartered in Dallas that claimed large portions of a customer database as trade secrets. The court found the claim overbroad in two compatible ways: the firm did not distinguish public from non-public information, and it never identified with specificity what within the database was the claimed secret. The court also noted there was no evidence the former employee used or disclosed the claimed secrets. The case did not involve AI. A narrow, well-protected set of secrets is easier to defend than a broad label applied to everything. That discipline also makes AI controls practical: a business can block a short list of tagged categories more reliably than “all confidential information.”

Timing

In Insulet Corp. v. EOFlow (Fed. Cir. May 28, 2026), the court set aside a verdict on limitations grounds because the plaintiff knew or should have known the critical facts more than three years before suing. AI-assisted monitoring of competitors' products and filings can reveal facts relevant to when a claimant discovered, or reasonably should have discovered, misappropriation, and should prompt timely investigation. An automated alert does not invariably start the clock: Insulet applied the Defend Trade Secrets Act discovery rule to its own facts and was not an AI-monitoring holding.

What Is Unsettled

To test an ascertainability argument, identify the claimed secret and the lawful sources available at the relevant time. Record the tool, inputs, access, output, effort and reproducibility. An output obtained by supplying the secret itself does not show reconstruction through proper means. The cited cases do not establish a categorical rule about AI-surfaceable information or public-chatbot uploads.

Identify the Information Before an Upload

A broad label such as confidential business data is a poor description of the claimed secret. Identify the specific pricing model, technical method, customer compilation or combination that has value from secrecy. Record which portion entered the tool and whether the output reproduces it or discloses its structure. DeWolff’s rejected database claim illustrates why a Texas claimant needs specificity. A prompt inventory can identify the material and its version without treating every ordinary employee question as a trade secret.

Read the Vendor Rights and Actual Access

Before uploading, compare the binding terms with the account configuration: permitted purposes, retention, support access, subprocessors, sharing and deletion. A training opt-out answers one question; it does not establish every secrecy measure. A protected enterprise workflow can differ from a consumer service that receives information under broader use rights. Keep the terms and settings that applied at the time, since a later screenshot does not establish the earlier permission. These are practical controls supporting a reasonable-measures analysis, not a new Texas AI contract checklist imposed on every business.

Test a Reconstruction Claim With Evidence

If a competitor says AI could reconstruct the information from public sources, ask which lawful inputs and prompts produced which output, at what effort and with what accuracy. Preserve the test and source materials. A description of what a model might do is weaker than a reproducible account of what it did. Comet assigns the DTSA plaintiff the statutory ascertainability burden; it did not adjudicate an AI reconstruction. The Texas statutory text and the actual means used remain part of a separate TUTSA analysis. A narrow classification system can protect the valuable information more credibly than labeling every file secret.