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 v. XP Power assigned the DTSA plaintiff the ascertainability burden; it did not decide an AI reconstruction issue. In DeWolff v. Pethick, the Fifth Circuit rejected an overbroad Dallas database claim without specific identification of the secrets. 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: 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 v. XP Power (9th Cir. July 14, 2026), the court vacated 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 decision does not bind Texas courts but is likely to be cited in DTSA cases here.

Identify Narrowly

In DeWolff, Boberg & Associates v. Pethick, 133 F.4th 448 (5th Cir. 2025), the Fifth Circuit affirmed summary judgment against a plaintiff that claimed large portions of a customer database as trade secrets without distinguishing public from non-public information. 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 speed discovery of misappropriation, which also starts the clock.

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.