Who Qualifies as an Inventor for AI-Assisted Patents in the United States?

Direct Answer to the Inventorship Question

Under U.S. patent law, an inventor must be a natural person who contributes to the conception of a claimed invention. An AI system, generative-AI service, software platform, or other machine cannot be named as an inventor, and using AI does not automatically allow a human applicant to claim the entire output as their own. For an AI-assisted invention, every natural person who contributes to the claimed subject matter must be identified, while technical contributors whose work relates only to an implementation or routine machine operation may be excluded. The controlling U.S. statute requires the inventor or inventors to make the prescribed oath or declaration, and current USPTO guidance applies a two-part test based on (1) contribution to the claimed invention and (2) significance of that contribution to the conception of the claimed subject matter.

Also worth reading: Can AI Be an Inventor, and How Should Patent Inventorship Be Determined for AI-Assisted Inventions in 2026? · How Does AI Patent Eligibility Review Work in the United States in 2026? · How far can a police officer search your car without probable cause or consent in the United States as of September 2026?

As of October 1, 2026, the core rule remains narrower than the phrase “inventorship attribution” may suggest. The relevant inquiry is not whether a person conceived an idea generally, used AI extensively, controlled prompts, or deserves credit as the project leader. It is whether that person made a significant contribution to the conception of at least one claimed invention and whether other people made comparable contributions to that same claimed subject matter. Consequently, an invention developed through AI can be patentable in an appropriate technological field, but the application still needs a legally acceptable human inventorship determination. Different jurisdictions apply different rules, so companies filing internationally should not assume that the U.S. approach automatically governs an overseas office.

How U.S. Inventorship Rules Work

The U.S. Supreme Court’s decision in Thaler v. Vidal held that an “inventor” under the patent statutes must be a natural person. That holding is consistent with 35 U.S.C. § 115, which directs the person claiming an invention to provide an oath or declaration stating that the applicant is the inventor. The statute does not allocate inventorship by employment title, effort spent, financial investment, authorship of a prompt, or ownership of the underlying algorithm. Patent rights may be assigned, but assignment changes ownership rather than inventorship: an assignee is not automatically entitled to be listed as the sole inventor if an employee or contractor supplied a patentable conception.

The USPTO’s 2024 AI-assisted-invention guidance distinguishes a person’s significant contribution to conception from a contribution made through routine or conventional means. Prompting an AI tool is therefore fact-specific rather than automatically qualifying or disqualifying. A user might contribute protectable subject matter by selecting the problem, supplying technical constraints, identifying a nonroutine arrangement of system components, or recognizing and evaluating an AI-proposed solution. A person who merely requests a desired result, receives an AI-generated candidate, verifies it, or asks a skilled colleague to implement the result ordinarily does not thereby become an inventor of every claim. The analysis remains claim-by-claim because an application may contain one correctly named inventor for one claim and additional inventors required for another.

The USPTO also follows an approach substantially derived from Pannu v. Iolab and Stanford v. Roche when allocating inventorship. Inventors need not be the sole source of every idea, and a person need not contribute to every feature of every claim. There must instead be a significant contribution to at least one claim, considering the claims as a whole, and no outstanding application or patent claim may improperly add a contributor who made no such contribution. This framework prevents an employer from disappearing from the application merely because it arranged the work, while also preventing employees or AI users from acquiring inventorship rights based only on effort.

Does Using Generative AI Create an Inventorship Problem?

Generative AI creates two distinct risks: the first is inventorship, and the second is ownership or unauthorized disclosure. If a developer uses AI only for ordinary activities—such as brainstorming, background research, editing grammar, or generating nontechnical text—the tool use may not affect inventorship if a human independently contributes the claimed conception. The risk changes when the human accepts the AI’s proposed technical solution without any meaningful human contribution to conception. In such a case, the developer may have failed to claim inventorship accurately, while naming the human merely because a natural person must be listed does not solve the problem.

Disclosure is a separate issue. Before an effective filing date in the United States, each inventor must satisfy the duty of candor to the USPTO and make a signed disclosure of information believed material to patentability. Information material to the conception of the claimed invention, potentially material prior art, and the status of testing may need to be disclosed. The USPTO has stated that using an AI tool may require disclosure of the category and function of the tool and how it was used, because that information could affect whether the claimed subject matter was conceived by the inventors or the disclosure could implicate prior-art or public-use issues. A record of prompts, outputs, model interactions, and contributor communications can therefore matter, although every prompt and response need not automatically become prior art.

Some AI systems interact publicly with a growing body of human content, but that does not mean every AI-generated output is necessarily prior art under every jurisdiction. Prior-art status depends on public availability, timing, the content actually disclosed, and applicable rules. Even where an output does not qualify as prior art, disclosure may still bear on inventorship, joint research, derivation, ownership, or evidentiary credibility. The defensible course is to document how the tool was selected and used, identify material content returned by it, and have patent counsel evaluate whether a duty-of-candor statement is warranted.

Human Inventor Versus AI “Inventor”: Practical Comparison

A useful comparison is also required among different human contributors. An engineer who conceives a specific technical mechanism may be an inventor, whereas a project manager who supplies business goals but contributes nothing technical to the claimed mechanism ordinarily is not. An engineer who contributes a limitation appearing only in claim 1 may qualify even if the claim also includes another person’s contribution, but that person’s name cannot be added merely because they supplied a supporting feature. Joint inventorship under U.S. law does not require equal effort; it applies when the relevant legal conditions are met, and the USPTO may inquire whether the joint inventors arose from a common plan or conception, although collaboration alone is not always enough.

A Practical Workflow for Patent Drafting Teams

The first practical step is to create a dated invention record before counsel begins drafting claims. The record should identify the problem, relevant technical background, each proposed solution, experiments, model names and versions used, the purpose of each AI interaction, and the humans who selected, modified, rejected, or combined proposed features. Keeping selected and rejected ideas is particularly useful because inventorship is assessed against the eventual claims, not against the commercial narrative the team later chooses. The goal is not to preserve every prompt mechanically, but to establish who made which significant technical contributions.

Next, counsel should map the technical contribution history to an early claim set. Inventor questionnaires and interviews should separately ask what each proposed inventor did and what each proposed non-inventor did. Questions such as “Were you on the project?” or “Did you own the code?” are inadequate. More informative questions ask whether the person proposed a feature, devised an operative relationship, selected among several alternatives for a claimed reason, or reduced a general objective to a specific workable solution. Material contributions should then be checked against each issued or allowed claim, including dependent claims, because amendments can change who must be named.

The filing itself should use a reliable inventor declaration executed by every properly identified inventor through the required USPTO process. If inventorship is uncertain, patent counsel may recommend a broader initial identification or an approach such as filing without an executed declaration when the system permits it, rather than making an unsupported factual assertion. An application lacking a properly executed declaration by the applicable statutory deadline may become abandoned. Under 35 U.S.C. § 116, the terminal disclaimer and petition provisions can create delay or forfeiture remedies where inventors are improperly omitted, but they are not substitutes for diligent review.

The team should also run an AI-specific disclosure review before filing. Counsel should determine whether the tool, its source or training materials, generated output, or third-party content is material to the claims or validity assessment. If necessary, the application can explain that AI was used for exploration or an intermediate step and identify the human intellectual contribution that produced the claimed invention. The review should be coordinated with prior-art searching because information about public AI interactions can change the significance and timing of references. Maintaining audit records after filing is prudent because inventorship and candor questions may arise during prosecution, assignment disputes, validity challenges, or litigation.

Common Mistakes and Consequences

The most common mistake is treating the prompt writer as the inventor without examining the claimed subject matter. Writing a detailed prompt may show intent and effort, but it does not necessarily show a significant contribution to conception. Another error is naming the founder, supervisor, or company because they directed the project or funded it. Ownership can arise by assignment or employment, but those facts do not establish inventorship. A third mistake is excluding an engineer who contributed a modest yet legally significant feature because that person did not conceive the entire system.

Incorrect inventorship can have procedural and substantive consequences. The USPTO may reject the oath or declaration, require correction, or refuse to recognize an improper inventor. If a required inventor is omitted, the omission may become incurable through ordinary disclaimer practice, while some naming errors may create statutory bars or affect standing. An erroneous statement is especially dangerous if made knowingly, although an incorrect oath does not necessarily mean every disputed patent claim is invalid. Counsel should avoid both underinclusive and overinclusive naming strategies and should analyze the actual contribution record.

Inventorship and eligibility should also not be confused. The inability to name AI as an inventor does not categorically prohibit AI-related patents, just as naming humans does not ensure that a claim satisfies the requirements for patentable subject matter. A software or AI claim must still satisfy eligibility requirements, novelty, nonobviousness, adequate written description and enablement, and disclosure standards. Similarly, describing an AI tool in a specification does not resolve ownership claims concerning training data, outputs, or collaboration with third parties. Those issues require separate contracts and, where necessary, litigation over contractual or statutory rights.

When Organizations Should Act and What It May Cost

Teams should address AI inventorship before filing, not after receiving a notice or enforcement demand. This matters especially for research-intensive companies, universities, startup collaborations, vendors using customer systems, and regulated sectors that need strong provenance records. An internal rule should apply when AI is used for ideation, model architecture, code generation, scientific design, optimization, or autonomous experimentation. Ordinary spelling correction poses less risk, but teams should not dismiss AI use merely because the output was ultimately edited; the legal question is the nature of the human contribution.

Timing is driven by the filing and disclosure obligations rather than by the duration of an AI project. Inventorship should be reviewed before a nonprovisional application is executed, provisional applications commonly should be filed within 12 months of an initial public disclosure, and a typical nonprovisional filing occurs around 18 months after a priority filing when no earlier filing is claimed. Those periods are not universal safe harbors: U.S. grace-period treatment is limited, foreign rules may be stricter, and a filing made before a public disclosure is generally the cleaner response.

Budgeting depends on complexity and available records. A minimal inventorship review for one modest application may cost roughly $750 to $2,500 when a patent attorney reviews clear contribution notes and an existing draft. A more involved review commonly ranges from about $2,500 to $10,000 or more where AI prompts, multiple experiments, contractor work, and competing claim interpretations must be reconstructed. Full patent drafting commonly begins around $8,000 to $20,000, while complex software, biotechnology, or foreign-filing matters can cost substantially more. Organizations should price provenance review as risk management, but avoid assuming that an expensive report cures missing human conception or an inadequate specification.

Key Takeaways for an AI-Related Filing Strategy

The definitive U.S. rule is that AI cannot be an inventor, and every qualifying human contributor to the claimed invention must be correctly identified. AI assistance is neither automatically fatal nor automatically attributable to the person who entered a prompt or received the output. The application must connect each named person to a significant contribution to conception, exclude persons whose contributions are routine or directed to implementation, and provide any disclosure required by the facts. International filings require a jurisdiction-specific review because natural-person and machine-inventor policies are not uniform, although treaty implementation does not remove every conflict.

For a company, the best practice is to preserve contribution records, separate conception from implementation, disclose material AI use, review inventorship against proposed claims, and revisit the issue after every material amendment. Such review should occur at least once before execution and again before allowance and issuance. The law does not reward the most senior claimant or the heaviest user of AI; it assigns inventorship based on legally relevant human conception. A careful record gives counsel the evidence to make that determination and helps the applicant defend both the inventor list and the patent’s validity.