Can AI Be an Inventor? AI Inventorship & Patentability in the USA, What Businesses Need to Know (2026)

The question of AI inventorship USA rules has moved from academic curiosity to boardroom urgency: every company using machine-learning tools in research and development must now decide how to file patents, whom to list as inventors, and what to disclose about AI’s role in the creative process. The Federal Circuit’s landmark decision in Thaler v. Vidal confirmed that only a natural person may be named as an inventor under the Patent Act, while the USPTO’s November 2025 Revised Inventorship Guidance reshaped examiner practice and applicant disclosure duties.

This guide delivers the practical playbook that in-house counsel, startup founders, and R&D leaders need, covering statutory rules, prosecution tactics, sample disclosure language, and a litigation-response framework, so your organisation can secure valid patent protection for AI-assisted innovations without stumbling into inventorship or validity traps.

Executive Summary & TL;DR for Counsel

The short answer is clear: AI itself cannot be listed as an inventor on a U.S. patent. The human inventorship standard applies. If a natural person made a significant intellectual contribution to the conception of the claimed invention, even when that person used AI as a tool, the invention is patentable and that person should be named as inventor. The critical compliance obligation is to disclose AI involvement when it materially affects how the invention was conceived or when it raises duties of candour toward the USPTO.

The USPTO’s Revised Inventorship Guidance for AI-Assisted Inventions, published in the Federal Register on November 28, 2025, rescinded several elements of the earlier 2024 framework and re-centred the analysis on traditional conception principles. Industry observers expect this re-alignment to reduce examiner uncertainty but to increase scrutiny of applicant statements about inventorship and AI use.

Legal Framework & Precedent, What the Courts and Statutes Say About AI Inventorship

Statutory Baseline, Patent Act Inventorship Language

Under 35 U.S.C. §100(f), an “inventor” is defined as the individual, or, for joint inventions, the individuals collectively, who invented or discovered the subject matter of the invention. Section 101 grants patents to whoever “invents or discovers any new and useful process, machine, manufacture, or composition of matter.” The consistent use of personal pronouns and the word “individual” throughout the Patent Act has been interpreted to require a natural person. Section 115 reinforces this by requiring each inventor to execute an oath or declaration, an act that only a human being can perform.

Thaler v. Vidal (Federal Circuit, No. 21-2347), Key Holdings

The Federal Circuit’s August 2022 decision in Thaler v. Vidal is the controlling precedent on AI inventorship in the USA. Stephen Thaler filed patent applications naming his AI system, DABUS, as the sole inventor. The USPTO refused the applications, and Thaler challenged the refusal in the Eastern District of Virginia, which upheld the agency’s position. The Federal Circuit affirmed, holding that the Patent Act unambiguously requires an inventor to be a natural person. The court examined the statutory text, legislative history, and Supreme Court precedent, concluding that Congress intended “inventor” to refer to human beings.

The practical takeaway is straightforward: naming an AI system as an inventor will result in rejection, and any patent issued with a non-human inventor would face serious validity challenges. The Supreme Court declined to hear the case, leaving the Federal Circuit ruling undisturbed.

Interaction with §115, Assignment, and Who May Own Rights

Although an AI cannot be an inventor, the patent rights that flow from a human inventor’s conception can be assigned to any legal entity, including a corporation whose employees used AI tools. Under 35 U.S.C. §261, patents are assignable by written instrument. The key is ensuring a clear chain of title from the natural-person inventor(s) to the assignee. Where AI tools were used under contract or licence, applicants must verify that neither the AI vendor’s terms of service nor any third-party licence restricts the assignability of inventions derived from AI output.

AI Inventorship USPTO Guidance and Federal Register Updates

February 2024 Guidance, Background

On February 13, 2024, the USPTO published its initial Inventorship Guidance for AI-Assisted Inventions, accompanied by illustrative examples. That guidance introduced a framework for evaluating whether a natural person’s use of AI rose to the level of inventive contribution. It emphasised that AI-assisted inventions are not categorically unpatentable and set out factors examiners should consider, including whether the human applicant provided a “significant contribution” to each claim.

The 2024 guidance was broadly welcomed by practitioners as a first attempt to bring certainty to an evolving area. However, questions persisted about how examiners would apply the “significant contribution” test in practice, particularly for inventions where the AI’s role was extensive and the human’s input consisted primarily of prompt engineering or parameter selection.

November 2025 Revised Inventorship Guidance, What Changed

On November 28, 2025, the USPTO published its Revised Inventorship Guidance for AI-Assisted Inventions in the Federal Register. The revised guidance rescinded several procedural elements of the 2024 framework and re-centred the inventorship analysis on traditional conception principles under existing case law. Instead of applying a standalone AI-specific test, examiners are now directed to apply the same inventorship evaluation they would use for any patent application, asking whether a natural person conceived of the claimed invention.

The revision also addressed the duty of disclosure. Applicants and their representatives have an ongoing obligation of candour under 37 C.F.R. §1.56. Where the use of AI is material to patentability, for example, because it raises questions about who actually conceived the invention or whether the specification enables a person of ordinary skill to reproduce the claimed subject matter, that information must be disclosed to the USPTO.

Practical Consequences for Examiners and Applicants

Under the revised guidance, examiners may issue office actions requesting additional information about the role of AI in the inventive process. Applicants should be prepared to provide detailed declarations or affidavits explaining human contributions. The likely practical effect will be a heightened focus on documentary evidence, lab notebooks, prompt logs, decision records, and versioned datasets, that demonstrates human conception. Applicants who fail to maintain such records risk protracted prosecution or, worse, post-issuance invalidity challenges grounded in improper inventorship.

Patentability of AI-Assisted and AI-Generated Inventions, §101, Novelty, Obviousness, Enablement

The question “Are AI-generated inventions patentable in the USA?” has a nuanced answer: yes, provided a natural person made a significant intellectual contribution to conception and the application satisfies all standard patentability requirements. The patentability of AI inventions does not turn on a single special rule, it turns on the same four pillars that govern every patent: eligible subject matter (§101), novelty (§102), non-obviousness (§103), and enablement/written description (§112).

§101, Eligible Subject Matter and the Abstract Idea Trap

Section 101 of the Patent Act defines patent-eligible subject matter, but the Supreme Court’s Alice/Mayo framework has created a notoriously unpredictable eligibility landscape for software and AI-related claims. Claims directed to an “abstract idea”, such as a mathematical algorithm or a method of organising human activity, must demonstrate “something more” that transforms the claim into a patent-eligible application. For §101 AI patents, this means that simply claiming the output of a machine-learning model without tying it to a specific, concrete technological improvement is likely to trigger a rejection.

The safest strategy is to claim the human-directed process, system, or apparatus that produces a specific technical result, rather than claiming the AI’s output in the abstract. Claims that recite specific hardware configurations, particular data-processing steps, or improvements to an existing technological process stand on firmer ground under current §101 jurisprudence.

Novelty & Obviousness, Documenting Who Conceived What

Novelty under §102 and non-obviousness under §103 require the applicant to establish that the claimed invention is new and would not have been obvious to a person of ordinary skill in the art. When AI tools are involved, the critical question is how to document conception. If a researcher instructs an AI model to explore a design space and the model returns a novel molecular structure, the researcher’s contribution, defining the problem, selecting the model, interpreting and validating results, must be carefully documented.

Prior art generated by AI also deserves attention. AI-generated publications, pre-prints, and datasets that are publicly accessible before the applicant’s filing date may constitute prior art under §102. Companies should conduct thorough prior-art searches that include AI-generated content repositories and preprint servers. Additionally, if multiple team members contributed to different aspects of an AI-assisted invention, joint inventorship under 35 U.S.C. §116 must be carefully analysed to avoid listing too many or too few inventors.

Enablement & Specification Issues for AI-Produced Artifacts

Under §112(a), the specification must enable a person of ordinary skill in the art to make and use the invention without undue experimentation. For AI-generated inventions, this requirement raises specific challenges. If the claimed invention depends on a particular trained model, the specification may need to describe the training data, model architecture, hyperparameters, and evaluation metrics in sufficient detail for reproducibility. Vague references to “a neural network trained on relevant data” are unlikely to satisfy the enablement standard.

Practitioners should also consider the written-description requirement. Where the AI’s contribution was to generate candidate solutions that a human inventor then selected and refined, the specification should clearly describe the selection criteria, validation steps, and any modifications the human made. Early indications suggest that examiners are paying closer attention to the enablement of AI-dependent claims, particularly in biotechnology and pharmaceutical applications where reproducibility has historically been scrutinised.

The interaction between enablement and trade-secret protection also warrants planning. Companies that wish to keep training data or model weights confidential must balance that desire against the requirement to disclose enough information for enablement. A patent strategy for AI inventions should address this tension at the outset, ideally before the first draft of the specification is prepared.

AI Inventorship Analysis, Who to List and Why

Natural Person Requirement and Joint Inventorship Principles

Under existing U.S. law, only natural persons may be named as inventors. Joint inventorship under §116 does not require each inventor to have contributed to every claim, but each named inventor must have made a contribution to the conception of at least one claim that is not insignificant in quality when measured against the full invention. The Pannu factors guide courts in evaluating joint inventorship: (1) the inventor contributed in some significant manner to the conception or reduction to practice, (2) the contribution is not insignificant in quality, and (3) the inventor did more than merely explain well-known concepts or the current state of the art.

When Human Prompts, Selection, or Curation Constitute Conception

The question of whether a human’s “prompting” of an AI system creates inventorship has no bright-line answer, it depends on the nature and specificity of the human input. Designing a detailed prompt that defines the problem space, sets constraints, and specifies evaluation criteria may well constitute conception if the resulting invention follows from those design choices. By contrast, entering a generic query into a large language model and selecting one of its outputs is unlikely to rise to the level of inventive contribution. Practitioners should document the iterative decision-making process: what prompts were tried, why certain outputs were selected, and what modifications the human made after receiving AI-generated candidates.

What NOT to Do

• Listing AI as an inventor. This will result in rejection under Thaler v. Vidal and the USPTO’s guidance.
• Overbroad attribution. Naming every team member who interacted with the AI tool, regardless of whether they contributed to conception, exposes the patent to invalidity challenges.
• Failure to document human contribution. Without contemporaneous records, an applicant may be unable to defend inventorship during prosecution or litigation.
• Ignoring vendor terms. Some AI-platform licences contain clauses that could affect ownership of outputs, review them before filing.

Practical Filing and Disclosure Checklist, How to Disclose AI in a Patent Application

Knowing how to disclose AI in a patent application is now a core prosecution skill. The following step-by-step checklist covers what to record, where to place disclosures, and how to tailor the approach by entity type.

• Step 1, Record-keeping. Maintain prompt logs, dated model versions, training-data manifests, and conception notebooks for every AI-assisted project from inception.
• Step 2, Identify human contributors. For each claim, document which natural person(s) conceived the inventive concept and how AI was used as a tool.
• Step 3, Third-party checks. Review the terms of service for any AI platform, API, or pre-trained model used. Confirm that no third-party ownership claim attaches to outputs.
• Step 4, Draft disclosure language. Prepare a concise statement for the specification describing the AI tool’s role and the human inventor’s contributions (see sample language below).
• Step 5, Place disclosures appropriately. Include the AI-use statement in the Detailed Description section of the specification. If material facts about AI use arise during prosecution, update the Information Disclosure Statement (IDS) or submit an applicant statement to the examiner.

Disclosure Obligations by Entity Type

Sample Disclosure Language

The following sample paragraphs may be adapted for use in a patent specification. They are illustrative and should be reviewed by qualified patent counsel before filing.

Where to Place Disclosures

The primary location for AI-use disclosure is the Detailed Description section of the specification. Background sections may also be appropriate for contextual statements about the AI tool’s role. If the applicant learns of material facts about AI involvement after filing, for example, that a key model was subject to restrictive licensing terms, the IDS or a supplemental applicant statement to the examiner is the proper vehicle for disclosure. There is no standalone USPTO form specifically for AI disclosure, so practitioners should integrate these statements into existing filing documents.

Prosecution & Patent Strategy for AI Inventions

Claims Drafting Tips

The most effective patent strategy for AI-generated inventions patent claims is to centre the claims on the human-directed method, system, or apparatus rather than on the raw AI output. Claim the steps a person performs, defining inputs, configuring model parameters, selecting and validating results, and implementing the solution, rather than claiming an abstract AI-generated result. Consider fallback dependent claims that recite specific human-implemented verification or modification steps, which can strengthen the argument for both §101 eligibility and inventorship.

IDS, Oath/Declaration, and Inventorship Corrections

Under 37 C.F.R. §1.63, each named inventor must execute an oath or declaration. Where inventorship needs to be corrected during prosecution, for example, because further analysis reveals that an additional team member contributed to conception, 37 C.F.R. §1.48 provides a mechanism for adding or removing inventors. Practitioners should perform the inventorship analysis early and revisit it before the notice of allowance to avoid post-issuance complications. Information disclosure obligations under §1.56 are ongoing: any material information about the AI’s role that comes to light during prosecution must be submitted.

International Filing Considerations

AI inventorship rules differ across jurisdictions. The European Patent Office, the UK Intellectual Property Office, and patent offices in Australia and South Africa have each adopted distinct approaches. Companies pursuing international filing strategies should prepare jurisdiction-specific inventorship analyses and disclosure documents. A failure to account for these differences can result in inconsistent patent portfolios or, in the worst case, invalidation in key markets. For a broader perspective on how to protect your intellectual property across borders, dedicated jurisdictional checklists are essential.

Responding to AI Inventorship Litigation and Validity Challenges, A Playbook

Common Attack Vectors

AI inventorship litigation is an emerging risk for patent holders. The most common challenge is an allegation of improper inventorship, that the true “inventor” was an AI system and the named human did not actually conceive the invention. Derivation proceedings under 35 U.S.C. §135 could also be relevant if a competitor alleges that the claimed invention was derived from AI-generated prior art rather than independently conceived. Enablement challenges targeting insufficient disclosure of training data, model architecture, or reproduction steps are another growing area of contention.

Corrective Measures

If an inventorship error is discovered after issuance, §256 allows correction of inventorship without invalidating the patent, provided the error occurred without deceptive intent. Reissue under §251 may be appropriate if the error is more fundamental. Reexamination (inter partes review or ex parte reexamination) is available to challengers who identify prior art or enablement issues related to AI use. Patent holders should be aware of estoppel risks, particularly that arguments made during prosecution about the human inventor’s role may limit claim scope or enforcement options later.

Evidence Checklist for Litigation Readiness

• Lab notebooks and conception records, dated, signed entries showing human decision-making at each stage.
• Prompt logs and model interaction records, timestamped records of AI queries, outputs reviewed, and selections made.
• Versioned model snapshots, preserved copies of the model, training data, and configuration at the time of invention.
• Internal communications, emails, Slack messages, or meeting notes reflecting human analysis and interpretation of AI outputs.
• Assignment and vendor agreements, executed copies confirming chain of title and absence of third-party ownership claims.

Conclusion, Recommended Immediate Actions for Businesses Navigating AI Inventorship USA Rules

The rules governing AI inventorship USA are now well-defined at both the judicial and agency level: AI cannot be an inventor, but AI-assisted inventions are fully patentable when a natural person contributes significantly to conception. The challenge for businesses is operational, building the documentation, disclosure, and filing processes that ensure every AI-assisted patent stands on solid ground.

Companies should take immediate, concrete steps. First, implement prompt-logging and conception-documentation protocols across all R&D teams that use AI tools. Second, update employment, contractor, and vendor agreements to include clear assignment clauses covering AI-generated output. Third, adopt standard disclosure templates, such as the sample language provided above, for inclusion in patent specifications. Fourth, conduct an audit of pending applications to verify that inventorship is accurate and AI use has been appropriately disclosed. Finally, engage qualified intellectual property counsel experienced in AI-related prosecution before making public disclosures or filing new applications. For further background on global IP strategy, the international intellectual property practice guide provides a useful starting framework.

Sources

1. USPTO, Revised Inventorship Guidance for AI-Assisted Inventions
2. Federal Register, Revised Inventorship Guidance for AI-Assisted Inventions (FR-2025-11-28)
3. Thaler v. Vidal, No. 21-2347 (Federal Circuit, August 5, 2022)
4. Sullivan & Cromwell, Memo on USPTO Guidance for AI-Assisted Inventions
5. Mayer Brown, Client Alert on USPTO Revised Guidance
6. Federal Register Notice 2025-21457, Justia Regulations
7. USPTO, FAQs on AI Inventorship Guidance