Key Points

• AI may be used to assist in the inventive process, but only human-developed aspects may be patented. Under current U.S. patent law, inventions developed with the help of artificial intelligence are eligible for patent protection as long as the claimed elements were conceived by a natural person. AI systems themselves cannot be named as inventors.
• The USPTO applies the same inventorship standard to AI-assisted inventions as to any other. Revised guidance issued in November 2025 confirms there is no separate legal test for AI involvement. The core question is whether the invention was developed by a human using AI as a tool or created autonomously by AI.
• Proper documentation and disclosure are critical to protecting AI-assisted patents. Inventors who use AI should keep detailed, contemporaneous records of their human contributions, be prepared for USPTO inquiries about AI involvement, and consider trade secret protection as an alternative when human conception is difficult to establish.

Artificial intelligence (AI) is transforming innovation. And it’s here to stay.

From identifying drug compounds and running simulations to generating inventions from prompts, AI now assists with all kinds of inventive work. The result is a growing portfolio of commercially valuable inventions that were developed in part with AI.

But can one still obtain a patent for such an invention?

The answer is nuanced and depends on how the AI was used during the inventive process.

Under current U.S. law, AI-assisted inventions are not categorically barred from patent protection. The key distinction is whether the invention was created autonomously by AI or developed by a human using AI as a tool.

Is There a Qualified Human Inventor?

A patentable invention must have a qualifying inventor who is a natural person. Two categories are relevant:

Category One: AI autonomously conceives a complete invention without meaningful human contribution

Under current law, such an invention may not be patentable because no natural person can be named as the inventor. All meaningful inventive aspects were generated by AI, and a computer cannot be an inventor.

Category Two: A natural person uses AI as a tool while conceiving an invention

Under this category, a natural person may have an inventive idea and use AI as a tool to research and develop it. The person may use AI like a scientist uses a microscope, calculator or research database.

Under this category, one or more natural persons may have developed one or more meaningful inventive aspects of an invention. As such, inventions conceived in this manner are not automatically barred from patent protection.

Since a qualifying human inventor is needed in order for an invention to qualify for patent protection, it is this second category, pertaining to inventions made by humans using AI as a tool, that will be discussed in more detail below.

Case Law: Thaler v. Vidal

The foundational case is Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022). Stephen Thaler filed two patent applications with the United States Patent and Trademark Office (USPTO), naming his AI system, DABUS, as the sole inventor. The USPTO rejected both patent applications because AI cannot be an “inventor” under the Patent Act, and Thaler appealed.

The U.S. Court of Appeals for the Federal Circuit affirmed, holding that the Patent Act unambiguously requires an “inventor” to be a natural person — that is, a human being. The court relied on the statutory definition: “the term ‘inventor’ means the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.” Because “individual” refers to a human being and Thaler named no human inventor, the applications could not proceed.

Thaler petitioned the U.S. Supreme Court for review. On April 24, 2023, the Supreme Court denied certiorari, leaving the Federal Circuit’s ruling in place. Unless Congress amends the Patent Act, AI systems cannot be named as inventors on U.S. patent applications.

The Federal Circuit expressly limited the scope of its holding, stating that it was “not confronted today with the question of whether inventions made by human beings with the assistance of AI are eligible for patent protection.” In other words, the decision did not address whether inventions developed through substantial human-AI collaboration are patentable, how much human contribution is necessary or whether Thaler could have been named as an inventor had he claimed sufficient involvement in the inventive process.

Current USPTO Guidance

On Nov. 28, 2025, the USPTO published revised guidance for AI-assisted Inventions in the Federal Register This guidance rescinded its February 2024 guidance, replacing it with a simpler, more straightforward approach. It states: “The same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used in the inventive process.” Thus, there is “no separate or modified standard for AI-assisted inventions.”

The guidance treats AI systems as “tools used by human inventors,” analogous to laboratory equipment, computer software and research databases. The same patentability analysis therefore applies: Did a human conceive the claimed invention?

So why was the 2024 guidance rescinded? The decision followed a change in presidential administration. The February 2024 guidance was issued pursuant to Executive Order 14110, which was signed by President Joe Biden on Oct. 30, 2023. On Jan. 23, 2025, President Donald Trump signed Executive Order 14179, directing agencies to review and revise prior AI policies to reduce regulatory burdens and promote U.S. competitiveness. The November 2025 guidance reflects that directive by returning to a streamlined, technology-neutral conception standard.

The Legal Standard: Conception

Under the revised guidance, the question is whether a natural person made the legally required contribution to the conception of the claimed invention. Conception is the touchstone of inventorship. As the Supreme Court explained, “[t]he primary meaning of the word ‘invention’ in the Patent Act unquestionably refers to the inventor’s conception.”

Conception is defined as “the formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention, as it is hereafter to be applied in practice.” A natural person who forms that idea is the inventor, even if AI provided assistance.

Conception is analyzed claim by claim. Each claim must have at least one qualifying human inventor who contributed to the conception of its subject matter. Different claims may have different inventors, and traditional principles of joint inventorship apply when multiple people contribute.

Sufficient and Insufficient Human Contribution

Although the February 2024 guidance was rescinded, its illustrative factors remain useful for assessing conception. Examples of potentially sufficient human contributions include:

• Formulating a specific technical approach or hypothesis rather than merely a general research objective.
• Devising or selecting features included in the patent claims.
• Making substantive technical refinements to an AI-generated proposal.
• Designing an essential component or building block of the claimed invention.
• Recognizing an AI output and developing it into the complete inventive concept.

Standing alone, the following activities likely do not establish inventorship:

• Giving the AI only a broad goal or objective.
• Owning or operating the AI system.
• Recognizing only afterward that an AI output is useful.
• Testing or building what the AI fully conceived.
• Claiming “intellectual domination” over the AI without contributing to the conception of the claimed subject matter.

The analysis is fact-specific. The key inquiry is whether the human formed “a definite and permanent idea of the complete and operative invention,” or whether the AI did so autonomously while the human only observed, selected or implemented the result.

If AI autonomously conceives the complete claimed invention and each human involved merely requests, observes, selects, tests or implements its output, no legally eligible inventor may exist. AI cannot be named because it is not a natural person, and no human can be named because none contributed to the conception. The invention could therefore be unpatentable despite being new, useful and nonobvious.

No U.S. appellate court has yet invalidated an issued patent on the ground that generative AI conceived it. The principal authorities remain agency guidance and the DABUS litigation.

Legislative proposals addressing patent eligibility have not resolved the AI inventorship gap. The Patent Eligibility Restoration Act of 2025 (PERA), introduced as S. 1546 in the 119th Congress on May 1, 2025, would replace judicially created exceptions to patentable subject matter — including those recognized in Alice, Mayo and Myriad — with narrower statutory exclusions. Although PERA could benefit AI-related software and machine-learning inventions by eliminating the abstract-idea framework, it would not address inventorship. It would neither change the rule that only natural persons can be inventors nor create a mechanism for protecting inventions autonomously generated by AI. No bill has been introduced to amend 35 U.S.C. § 100(f) to include or accommodate AI systems as inventors.

Duties Owed to the USPTO

If AI use does not automatically defeat patentability, must applicants disclose it to the USPTO?

In April 2024, the USPTO issued separate guidance on using AI-based tools in practice before the Office. It clarified that the existing duty of candor and good faith applies to AI use.

Applicants generally need not disclose AI use in drafting claims, conducting prior art searches or developing an invention. Disclosure is required if AI use is material to patentability. For example, AI use may be material to patentability if contributions and embodiments attributed to the named inventor(s) were, in fact, generated by AI.

Relying on an AI tool alone does not satisfy the “reasonable inquiry” required by 37 C.F.R. § 11.18(b). Practitioners must independently verify accuracy rather than rely on AI-generated output.

Examiners may also request information under 37 C.F.R. § 1.105 when they reasonably believe a party has information necessary to examine the application. This may include cases in which evidence suggests that a human did not significantly contribute to an AI-assisted invention even though the application names natural persons as inventors. Applicants should be prepared to substantiate human contribution.

Risks of Improper Inventorship

Incorrect inventorship can have serious legal consequences and may render a patent unenforceable or invalid.

Knowingly naming an incorrect inventor may support a claim of inequitable conduct (essentially, fraud on the USPTO). Because the inventor’s oath or declaration is submitted under the penalties prescribed in 18 U.S.C. § 1001, misstatements about who conceived the invention can create enforcement problems and ownership disputes, particularly when the chain of inventive contribution is unclear. Discovery about AI use in later litigation may further undermine the patent holder’s position.

Practical Guidance: Documentation

Those who develop inventions with AI should maintain thorough, contemporaneous records showing human contributions sufficient to establish conception. Dated records should identify:

• The human’s initial technical concept or hypothesis.
• The exact prompts and parameters supplied to the AI system.
• Which portions were generated by AI and which were contributed by the human.
• Why AI outputs were rejected or modified.
• Human-developed changes incorporated into each claim.
• When the human recognized the complete and operative invention.

Create these records contemporaneously. In a USPTO inquiry, an inter partes proceeding or patent litigation, they may help support the inventorship designation.

Trade Secrets as an Alternative

When no human can satisfy the inventorship requirement, trade secret law may offer an alternative. Trade secrets require no showing of inventorship or public disclosure, can last indefinitely while secrecy is maintained and require no filing or registration fees. Trade secret protections may be available for AI-generated information if the holder takes reasonable measures to maintain secrecy.

Trade secret protection is governed by state versions of the Uniform Trade Secrets Act and, at the federal level, by the Defend Trade Secrets Act of 2016. Where human conception is difficult to establish, trade secret protection merits consideration as an alternative or complement to patent protection.

Trade secret protection also has important trade-offs. Unlike a patent, it confers no exclusionary right and a competitor may independently develop or reverse engineer the same innovation. Lawful discovery by a third party ends the protection.

Trade secret protection also depends on maintaining secrecy. Publication, an employee’s departure, inadequate security measures or another inadvertent disclosure may permanently destroy protection.

Trade secret protection is strongest when the innovation is not readily ascertainable from the final product and the holder can maintain robust confidentiality controls.

International Dimensions

Although this article focuses on U.S. law, other major patent offices have reached similar conclusions on AI inventorship, with some differences in approach.

At the European Patent Office (EPO), the Legal Board of Appeal held in J 0008/20 that “[a] machine is not an inventor within the meaning of the EPC,” dismissing Thaler’s DABUS appeal. In T 0528/25 (decided Feb. 5, 2026), a Technical Board of Appeal confirmed that the EPC does not bar patent protection merely because AI helped develop an invention, even though AI cannot be designated as an inventor. The Board observed that how an invention was made “apparently plays no role” in the European patent system. A natural person may therefore be designated for an AI-assisted invention, but the designation must identify a person with legal capacity and cannot simultaneously assert that AI, rather than that person, conceived the invention.

In China, the National Intellectual Property Administration (CNIPA) issued its Guidelines for Patent Applications for AI-Related Inventions (Trial Implementation) on Dec. 31, 2024. The guidelines identify four categories, including “inventions made with AI assistance” (substantial human contribution) and “AI-generated inventions” (no substantial human contribution). Like the United States and the EPO, CNIPA requires inventors to be natural persons, not AI systems. It reasons that only “civil subjects” under Chinese civil law may enjoy the civil rights of inventors, and AI systems are not currently civil subjects.

The emerging international consensus is that AI cannot be an inventor, but AI-assisted inventions are not categorically barred from patent protection if a natural person can be identified. Practitioners filing in multiple jurisdictions should ensure inventorship designations are consistent and supportable before each relevant patent office.

Conclusion

AI use in the inventive process does not automatically disqualify an invention from patent protection. AI cannot be named as an inventor; at least one named human inventor must have genuinely conceived the claimed invention — not merely prompted an AI system, selected its output or implemented its proposals. The traditional inventorship standard applies — there is no special or modified AI test.

Those developing inventions with AI assistance should document human contributions contemporaneously, prepare for USPTO inquiries about AI involvement and consider trade secret protection when human conception is difficult to establish. Because this area is evolving, practitioners should monitor agency and legislative developments.

For more information, please contact Anthony J. Noonan at 609.895.6621 or anoonan@foxrothschild.com or another member of Fox Rothschild’s national Intellectual Property Department.