How Do the USPTO’s 2026 AI Inventorship Rules Affect Human Inventors?

Direct Answer: AI Cannot Be Named an Inventor Under Current USPTO Rules

As of September 25, 2026, the United States Patent and Trademark Office generally requires a natural person to be named as the inventor on a U.S. patent application. Artificial intelligence systems, including generative models, machine-learning systems, autonomous agents, and conventional algorithms, cannot be patent inventors merely because they produce a claimed invention. If an AI-generated output contributes to the conception of an invention, the application must identify the natural person who made the claimed invention and explain the role of the AI, if material, without naming the AI as an inventor.

Also worth reading: How Does AI-Assisted Patent Filing Work in 2026, and What Are the Inventorship Rules? · What is the definitive status of the USPTO AI inventorship 2027 rulemaking and how should IP teams prepare? · What are the mandatory USPTO patent filing requirements for foreign inventors and companies?

The USPTO’s February 13, 2024 guidance on AI-assisted inventorship is not a special registration system for AI-created patents. It explains how existing inventorship and oath rules apply when people use AI during research, drafting, testing, problem-solving, or invention development. A named inventor is not required to have conceived every technical feature personally, but that person must have made a significant contribution to at least one claim and must not merely have arranged for someone else to do the inventive work. The practical standard is therefore claim-specific: inventorship follows the contribution to the claimed subject matter, not the amount of AI used in the overall process.

The most important filing consequence is that patent applications should not list “ChatGPT,” “Claude,” “DABUS,” a company, or an internal automated system as an applicant-side inventor. Human legal entities may own assigned patent rights, but only natural persons are eligible inventors. A human applicant may own the disclosure, yet ownership does not cure an incorrect inventorship declaration. Counsel should resolve questionable inventorship before filing, while evidence and technical records should be preserved throughout development.

How USPTO AI-Assisted Inventorship Rules Work

The USPTO applies the same foundational concept used in Thaler v. Vidal and later judicial treatment of machine inventorship: inventorship is reserved for a natural person. Patent eligibility, by contrast, is a separate question. An invention must also satisfy the statutory requirements for patentable subject matter, including utility, novelty, nonobviousness, and eligibility under 35 U.S.C. § 101. A technically produced result is not automatically patentable merely because a human supplied inputs, and a patentable arrangement does not permit an AI to be named as its inventor.

USPTO guidance also distinguishes an AI tool from a human inventor. Generative AI is generally treated as an ordinary research or drafting tool, much like a calculator, search engine, simulation package, or specialized database. Merely prompting a model to propose a solution ordinarily does not, without more, establish that the person who conceived the claimed invention. The human must determine whether the proposed idea works, contribute the claimed conception, and be willing to claim the invention. The USPTO’s 2024 clarification effectively rejected the idea that a person becomes the inventor of every output generated through sufficient prompting.

At the same time, the rules do not require a human to avoid AI or to demonstrate that every idea originated without automated assistance. If a human contributes a specific technical solution, selects a promising AI-generated concept, and provides the insight that turns that concept into the claimed invention, the human may be an inventor. Inventorship is not decided by counting prompts, tokens, model runs, or hours. It is decided by identifying who caused the conception of the claimed features and whether that contribution was significant rather than merely administrative or ministerial. This is why technical role assignment should be evaluated person by person, claim by claim.

Human Contribution, Claim-by-Claim Analysis

USPTO practice requires a person to have made a significant contribution to the conception of at least one claim. Courts have commonly assessed contribution by comparing the claimed features with the inventor’s contribution to conception, not with the contribution to every aspect of the project. A person who supplies background information, selects a commercially important target, or tells a team to build something generally does not become an inventor solely because of that activity. A person who conceives a particular mechanism needed to solve the problem may qualify even if other contributors supplied different claims or embodiments.

The correct unit of analysis is therefore the claim, not the model or the entire patent. An invention involving three independent technical concepts may require careful allocation among multiple natural persons based on what each person actually conceived. If one person devised a specialized sensor arrangement but a different person devised a control system, their inventorship may not be identical. Counsel should map each claim and material embodiment to named individuals, identify the relevant technical contribution, and check whether the inventors are correctly ordered or jointly listed under the applicable U.S. inventorship rules.

Prompting can be evidence of contribution, but the number or length of prompts does not itself establish inventorship. A prompt requesting a known solution can fail to show conception, while one prompt containing a specific technical problem, proposed mechanism, and selection rationale can materially contribute to an invention. Recordkeeping should therefore capture the human’s hypotheses, rejected alternatives, calculations, test results, selection decisions, and any modifications made to AI suggestions. This creates a defensible explanation without treating conversational language as conclusive proof. The USPTO ultimately examines what the application claims and what the evidence shows about the human’s role.

What the USPTO Guidance Does Not Decide

The AI inventorship rules do not determine whether an output is novel or nonobvious. Inventorship asks who conceived the claimed invention; patentability asks whether the claim is legally allowable and distinguishable from prior art. An AI-generated output can be produced by a human inventor and still be obvious, abstract, unsupported, or anticipated. Conversely, correcting a naming error does not resolve whether the underlying invention qualifies for patent protection.

Nor do these rules establish a general right to practice AI. Someone may help invent a process using a lawful tool without infringing, infringe a third party’s patent, or violate contractual, trade-secret, data-use, and licensing terms. A team’s use of a public model can also raise issues involving access restrictions, confidential information, ownership of inputs and outputs, and employment duties. Those questions should be addressed through tool selection, vendor diligence, contracts, and internal controls rather than through the patent inventorship declaration.

Copyright treatment is also separate. The U.S. Copyright Office’s human-authorship analysis concerns works protected by copyright, while the USPTO’s framework concerns patent inventorship. A result could lack copyright protection in a particular jurisdiction yet still be capable of contributing to a patent application if a natural person satisfies the inventorship requirements. Patent disclosure and copyright ownership are distinct legal treatments. Decisions concerning one should not be treated as automatically answering the other, and international patent offices may apply different rules to AI-related inventorship and entitlement questions.

Practical Steps for a U.S. Patent Filing

The first practical step is to create a contemporaneous record of the human contribution. Engineers, researchers, and product teams should identify the technical problem, preserve their notes, and distinguish ideas proposed by people from concepts generated by an AI tool. A dated laboratory notebook, design document, issue report, experiment, sketch, or technical email can show how a person recognized and solved the relevant problem. AI transcripts may be retained when relevant, but the company should not assume that the transcript proves inventorship either way.

The second step is to assess contribution before choosing the inventor list. For each proposed claim, counsel should identify the person or people who conceived the claimed subject matter and exclude persons who provided only general instructions, project management, routine testing, or implementation. If the records are uncertain, technical interviews and supplementary factual statements may be needed. Inventorship can be corrected during prosecution, but correction is not a substitute for a reasonable investigation before execution, and knowingly or recklessly naming the wrong inventor may create legal problems.

The third step is to prepare the specification in human-verifiable terms. The application should describe an invention that the applicants understand, enable, and have support for in the written description. AI output should be reviewed for technical accuracy, completeness, consistency, and alignment with the actual human contribution. The specification should not contain unsupported assertions about mechanisms, dimensions, performance, or tests merely because a model generated plausible language. Search and drafting assistance can improve efficiency, but neither the USPTO nor a court should have to guess which material was invented and which was merely suggested.

The fourth step is to investigate before filing, not after a dispute arises. Teams that lack internal technical records can incur disproportionate costs reconstructing who conceived what. Decisions that may require factual investigation include changing an inventor after execution, including an employee as an inventor, assigning rights from several contributors, or filing where AI performed much of the inventive step. The relevant investigation should be tailored to the application’s technical facts rather than relying on a universal statement that all AI users are—or are not—inventors.

Organizations sometimes want a registry model that automatically attributes a result to a model, dataset, or AI agent. Such a system could be useful for internal provenance, licensing, audit, or computational reproducibility, but it is not equivalent to a USPTO patent filing. Inventorship is a legal status attached to a natural person’s contribution, while internal attribution may reflect whichever system generated a candidate output. A registry can record prompts, model versions, datasets, contributors, and ownership conditions, but it should not represent itself as naming a legally valid patent inventor unless the relevant person has satisfied the USPTO requirements.

Alternative patent offices have also approached AI inventorship differently, and international filings should not be assumed to receive identical treatment. The legal analysis may depend on the jurisdiction, the identity of the applicant, and the wording of the claims. Companies with global ambitions often need coordinated advice rather than a single U.S.-style form completed everywhere. Domestic counsel should confirm the law in each target jurisdiction, while U.S. counsel should continue to follow USPTO practice and current examination guidance.

Ownership is another distinct layer. A natural person may be the inventor while a company owns the application through assignment. Inventorship ordinarily cannot be assigned away, and naming the assignee as the inventor is not permitted merely because the assignee funded the work. Joint ownership may also carry consequences for licensing and enforcement, so teams should not confuse an employment agreement with a determination of inventorship. For B2B intellectual-property operations, a clean record should therefore separate four fields: human inventors, applicants or assignees, inventors’ rights, and the role of software and AI systems.

Common Mistakes and When Organizations Should Act

A common mistake is treating every AI user as the inventor. Another is assuming that the person who entered the winning prompt must be listed even when a specialist made the operative technical contribution. Teams also err by naming a team, employer, model, or autonomous agent; these choices can trigger formalities issues or questions about the application’s inventorship declaration. The opposite mistake is assuming that extensive AI use makes filing impossible. A human can potentially use AI substantially and still satisfy the inventorship requirement, provided the human conception is real, significant, and reflected in the claims.

Another error is confusing confidentiality with inventorship. Keeping details from the USPTO may create a problem if the written description is deficient or the inventor declaration is inaccurate. Redaction may be appropriate in limited contexts, but public patent applications generally must disclose enough information to support the claimed invention. Open-source licenses, model terms, and data restrictions should be reviewed separately. A permitted research tool can still generate prior art, and a confidential collaboration can create contractual or trade-secret exposure even before anyone files an application.

Organizations should act early when AI materially participates in conception, when multiple contributors are involved, or when inventorship cannot be reconstructed from records. A reasonable trigger is any proposed filing that includes an AI-generated architecture, algorithm, formula, material composition, or technical mechanism—not only a fully autonomous system. The review can be scaled to the project, but there is no bright-line number of prompts, model queries, or dollars spent below which inventorship concerns disappear. A project using a standard model for search and spelling should not receive the same scrutiny as a system that proposes the claimed invention, but the intensity of review should follow the actual technical contribution.

Cost depends on the dispute. A public information request and organized technical records may require little beyond internal time, while a contested inventorship investigation can involve outside patent counsel, technical experts, declarations, prior-art analysis, and prosecution amendments. Patent office fees do not include this factual work. In a complex multi-inventor AI project, budgeting time for a contribution review before filing can be more economical than correcting inventorship or defending a declaration later. A registry SaaS platform can organize dates, artifacts, assignments, and review status, but it is an operational aid rather than legal advice or a substitute for a USPTO determination.

Bottom-Line Guidance for Counsel and Product Teams

The USPTO’s current position is straightforward: a natural person must satisfy the inventorship requirement, and an AI cannot be named as the inventor. People can use AI tools, including generative systems, in developing inventions, but the application and supporting evidence should show that the named human or humans contributed significantly to the conception of the claimed subject matter. Naming the person who commissioned a machine-generated idea is not enough; prompt count and software sophistication are not decisive either.

The prudent response is not to ban AI or to claim that every human becomes an inventor of everything the system outputs. Teams should document the human problem-solving process, evaluate each claim, verify technical disclosure, check contractual rights, and use U.S. counsel before execution when contribution is uncertain. For B2B intellectual-property operations, this means treating AI provenance, human inventorship, ownership, confidentiality, and patentability as separate records with separate decisions. That discipline helps a company file accurately while avoiding the costly mistake of allowing a useful software system to stand in for a person who did not make the required inventive contribution.