AI Patents and Trade Secrets in Texas

Short Answer

Only a natural person can be an inventor on a U.S. patent. In November 2025 the USPTO rescinded its 2024 AI inventorship guidance in its entirety and replaced it with a simpler framework: AI systems are tools, analogous to laboratory equipment, and inventorship is judged under the ordinary conception standard. The joint-inventorship "significant contribution" factors now apply only among multiple human collaborators. On the trade secret side, the Texas Uniform Trade Secrets Act protects information of independent economic value that the owner keeps secret through reasonable measures. Pasting prompts into a public consumer AI tool can destroy the "secret" before any misappropriation claim begins: in Trinidad v. OpenAI (N.D. Cal., January 5, 2026), a federal court held, on a motion to dismiss decided on the complaint’s pleaded facts, that developing claimed trade secrets inside a public AI tool was voluntary disclosure to a party with no duty of confidentiality, so the secrecy element failed at the threshold.

Which Laws Apply

• Federal: Patent Act inventorship requirements (35 U.S.C. §§ 100 and 115); USPTO inventorship guidance; the Defend Trade Secrets Act (DTSA), 18 U.S.C. § 1836 et seq.
• Texas: Texas Uniform Trade Secrets Act (TUTSA), Civil Practice and Remedies Code chapter 134A. TRAIGA has no patent or trade secret provisions; these questions are answered by general law applied to an AI use.

Who Can Be an Inventor: the November 2025 Revised USPTO Guidance

The background rule is settled. In Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), the Federal Circuit held that only natural persons may be listed as inventors; the Supreme Court declined review. An AI system cannot be an inventor or a joint inventor, and listing one invites rejection.

The analytical framework around that rule changed in late 2025. On February 13, 2024, the USPTO had issued "Inventorship Guidance for AI-Assisted Inventions" (89 FR 10043), which borrowed the joint-inventorship factors of Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998) and applied them to assess whether a single human's contribution to an AI-assisted invention was "significant." On November 28, 2025, the USPTO published "Revised Inventorship Guidance for AI-Assisted Inventions" (90 FR 54636), which rescinds the 2024 guidance in its entirety.

The revised framework, as written in the Federal Register notice:

• One standard, no AI carve-out. The same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used in the inventive process. There is no separate or modified standard for AI-assisted inventions.
• AI is a tool. AI systems, including generative AI and other computational models, are instruments used by human inventors, analogous to laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process. They may provide services and generate ideas, but they remain tools used by the human inventor who conceived the claimed invention.
• Single natural person plus AI: the conception question. When one natural person creates an invention with AI assistance, the inquiry is whether that person conceived the invention under the traditional conception standard: the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention.
• Multiple natural persons plus AI: ordinary joint-inventorship rules. When multiple humans collaborate with AI assistance, the traditional joint inventorship principles apply among the humans, including the Pannu factors. The Pannu factors do not apply to the human-AI relationship, because AI systems are not persons and cannot be joint inventors, so there is no joint inventorship question to analyze.
• Scope. The guidance applies to utility, design, and plant patent applications. Benefit and priority claims require natural-person commonality: a U.S. application may not list a non-natural person as a joint inventor to claim priority.

Practical consequence for Texas R&D teams: name the humans who conceived each claimed invention, and keep contemporaneous records of who formed which inventive concept. Where only a prompt went in and the model supplied the complete inventive concept, the human has a conception problem, not a "significant contribution" problem, because that test no longer exists for single-inventor AI-assisted cases. And because the guidance binds examiners, not district courts, the outer boundary of conception when AI does heavy analytical lifting remains an open Federal Circuit question.

TUTSA Secrecy Elements and the Public AI Tool Problem

TUTSA defines a "trade secret" in § 134A.002(6) as information in any form, including business, scientific, technical, economic, or engineering information and any formula, design, prototype, pattern, plan, compilation, program, code, device, method, technique, process, procedure, financial data, or customer or supplier list, if two elements hold: (A) the owner has taken reasonable measures under the circumstances to keep the information secret, and (B) the information derives independent economic value, actual or potential, from not being generally known to, or readily ascertainable by, another person who can obtain economic value from its disclosure or use.

Reasonable measures is a threshold element. A plaintiff who cannot show it survives loses before the court reaches misappropriation. Public consumer AI tools threaten that element directly: inputs may be logged, reviewed, used for training or abuse monitoring, and shared with vendor personnel, under terms the user accepted but rarely negotiated.

Trinidad v. OpenAI, No. 25-cv-06328-JST (N.D. Cal. January 5, 2026), is the first decision squarely on point. Pro se plaintiff Rebecca Trinidad alleged that OpenAI took AI "frameworks" she developed through ChatGPT and incorporated them into its products. Judge Jon S. Tigar granted OpenAI's motion to dismiss with prejudice. The trade secret holdings, taken from the signed order:

• Developing the claimed "protocols and frameworks" through ChatGPT "would have required her to voluntarily share the information she now alleges is part of her 'trade secrets' with OpenAI," a party under no obligation of confidentiality. Disclosing a claimed secret to a party under no obligation to protect it extinguishes the property right, citing Ruckelshaus v. Monsanto Co., 467 U.S. 986, 1002 (1984). The DTSA reasonable-measures element failed at the threshold.
• Ownership is not secrecy. OpenAI's Terms of Use assigned output ownership to the user, but the court held that ownership does not satisfy the secrecy requirement of 18 U.S.C. § 1839(3): "whether she owned those ideas notwithstanding the disclosure is not relevant."
• Consent stands either way. Her consent to the disclosure stood "whether or not [her] consent is enforceable as a contractual matter," defeating the argument that the terms were an unconscionable contract of adhesion.

Trinidad is non-binding in Texas, but TUTSA mirrors the DTSA's reasonable-measures element, so the reasoning travels. Note the decision's limits: it was decided on a motion to dismiss, taking the complaint's pleaded facts as true, so it shows how the reasonable-measures defense plays at the pleading stage rather than how every court will rule on a full record. Warner v. Gilbarco Inc. (E.D. Mich., February 2026) is sometimes mentioned alongside Trinidad, but it answers a different question: the court held litigation materials a pro se plaintiff prepared with ChatGPT stayed protected as work product, because work-product waiver requires disclosure to an adversary and generative AI programs "are tools, not persons." That is a waiver analysis under Federal Rule of Civil Procedure 26(b)(3), not a ruling on trade-secret secrecy, so the two decisions do not form a split on the reasonable-measures question. The practice point stands on Trinidad alone: do not let a client's trade secret position depend on a court taking a friendlier view of the secrecy element.

The defensive move is contractual, and it mirrors the Pieces lesson from the health care side: use enterprise tiers with no-training commitments, confidentiality provisions, and data processing addenda, which courts have long accepted as reasonable measures when sharing secrets with a vendor under an NDA. Even then, note that enterprise tiers typically retain data temporarily for abuse monitoring and permit limited vendor personnel access, so the residual risk should be documented as an accepted one, not assumed away.

Reverse Engineering as Proper Means, "Unless Prohibited"

Civil Practice and Remedies Code § 134A.002(4) defines "proper means" as "discovery by independent development, reverse engineering unless prohibited, or any other means that is not improper means." § 134A.002(5) defines reverse engineering as the process of studying, analyzing, or disassembling a product or device to discover its trade secrets. "Improper means" under § 134A.002(2) includes theft, bribery, misrepresentation, breach or inducement of breach of a duty to maintain secrecy or limit use, and espionage through electronic or other means.

The "unless prohibited" phrase is the drafting lever the statute hands to Texas businesses. Reverse engineering is a proper means of discovering a trade secret by default, but a contractual prohibition changes that: license agreements, terms of use, NDAs, and employment agreements that bar reverse engineering, disassembly, or decompilation can convert what TUTSA would otherwise treat as proper discovery into a breach. For AI products, that means the license terms a Texas company puts on its own models, APIs, and outputs are themselves a trade secret protection measure, and the terms it accepts from AI vendors determine what it may lawfully learn from their products.

Illustrative Example (Hypothetical)

An Austin startup's engineers paste a proprietary pricing algorithm into a public chatbot to debug an edge case, then file a patent application naming the lead engineer as sole inventor and later sue a competitor that ships a similar feature. Two independent failures follow. On the patent side, if the model's output supplied the complete inventive concept and the engineer's contribution was the prompt, the application has a conception problem under the November 2025 guidance, and the named inventor may not be an inventor at all. On the trade secret side, the TUTSA claim fails at reasonable measures: the algorithm was voluntarily disclosed to a party with no duty of confidentiality, exactly the Trinidad pattern, and the vendor's assignment of output ownership does not restore secrecy. Had the startup used an enterprise tier with a no-training commitment and confidentiality terms, documented the debugging as an authorized use, and kept conception records showing which human formed the inventive concept, both positions would look materially different.

What Is Unsettled

Whether consent to consumer AI terms of use defeats reasonable measures in every case or turns on the specific data-use terms; whether enterprise-tier temporary retention for abuse monitoring is consistent with reasonable measures; and the outer boundary of conception when AI performs the heavy analytical lifting in a multi-claim invention.

Questions to Ask

• Which humans conceived each claimed invention, and do contemporaneous records show who formed which inventive concept?
• If only a prompt went in, did a natural person form a definite and permanent idea of the complete invention, or did the model?
• Does any claim name or depend on an AI system as inventor, including through a foreign priority claim?
• Before pasting proprietary material into an AI tool: does the tier used carry a no-training commitment, confidentiality provisions, and a data processing addendum, or is it a public consumer tier?
• Do vendor terms permit training, human review, or retention beyond abuse monitoring, and has that residual risk been documented?
• Do the company's own licenses and terms prohibit reverse engineering, disassembly, and decompilation of its AI products?
• Are employee AI-use policies specific about which information may go into which tier, with the Trinidad reasonable-measures consequence stated plainly?