AI-Assisted Inventions under US Patent Law: Who Counts as the Inventor?

Key points

The current US position is narrow and clearer than much of the public debate suggests:

• Statutory rule: the Patent Act defines an inventor as the “individual” or “individuals” who invented or discovered the claimed subject matter. The Federal Circuit held in Thaler v. Vidal that this means a natural person. An AI system cannot be named as a sole inventor or joint inventor.
• Limit of the court’s holding: Thaler involved two applications that named DABUS as the sole inventor. The court expressly left open the patentability of inventions made by human beings with AI assistance.
• Current agency position: the USPTO’s November 2025 guidance treats AI as a tool and applies the ordinary law of conception. It creates no separate inventorship standard for AI-assisted work.
• Material change since 2024: the USPTO rescinded its February 2024 guidance in its entirety. Its earlier use of the Pannu joint-inventorship factors to assess a single human working with AI is no longer the agency’s approach.
• Separate legal questions: a correctly identified human inventor does not make a claim patentable. Eligibility, novelty, non-obviousness, disclosure and other statutory requirements must still be satisfied.

Legal-information notice: This article explains public US authorities as at 14 August 2026. It is not legal advice. Inventorship depends on the claims, the actual course of development and evidence that may not be visible from a final AI output or prompt log.

Current legal landscape

The Patent Act requires a natural person

Statutory rule. 35 U.S.C. § 100(f) defines “inventor” as the individual, or the individuals collectively for a joint invention, who invented or discovered the subject matter. Section 100(g) uses the same individual-centred language for a joint inventor. Section 115 requires each inventor or joint inventor to execute an oath or declaration, subject to specified substitute-statement provisions.

Precedential holding. In Thaler v. Vidal, Stephen Thaler listed his AI system, DABUS, as the sole inventor of a “Neural Flame” and a “Fractal Container”. He maintained that he had not contributed to their conception. The Federal Circuit affirmed the USPTO’s rejection of the applications because the statutory term “individual” refers to a human being. The decision did not grant rights to AI, decide who owns AI-generated output, or determine whether a human-assisted invention satisfies the other patentability requirements.

The Supreme Court docket records that certiorari was denied on 24 April 2023. A denial of certiorari is not a ruling on the merits and does not convert the Federal Circuit’s reasoning into a Supreme Court holding. It left the Federal Circuit judgment in place.

The November 2025 USPTO guidance changed the agency framework

USPTO examination guidance. On 28 November 2025, the USPTO published revised guidance at 90 FR 54636. It expressly rescinded the February 2024 guidance in its entirety. This matters because the superseded document had adapted the Pannu joint-inventorship factors to ask whether a natural person made a “significant contribution” to an AI-assisted invention.

The revised guidance says there is no separate or modified standard for AI-assisted inventions:

• one natural person working with AI is assessed under the traditional conception standard;
• several natural persons working with AI are assessed under ordinary joint-inventorship principles, including Pannu, as between those people;
• an AI system remains an instrument rather than an inventor; and
• an application naming an AI system or another non-natural person as an inventor or joint inventor should face rejection under 35 U.S.C. §§ 101 and 115, or other appropriate action.

The distinction between a court holding and agency guidance should remain visible. Thaler authoritatively resolved whether an AI system may be named as the inventor under the present statutory text. The 2025 guidance directs USPTO examination of human-AI workflows. It is persuasive evidence of current agency practice, but it is not legislation and does not report a later precedential decision applying the conception standard to a fully contested human-and-AI record.

Conception remains the central question

The revised guidance draws on established Federal Circuit doctrine describing conception as the point at which an inventor possesses a definite and permanent idea of the complete and operative invention. A general objective or research plan is insufficient. The inquiry asks whether the claimed solution was sufficiently settled in the natural person’s mind that ordinary skill, rather than extensive further research, would be enough to reduce it to practice.

This is a claim-focused and fact-intensive inquiry. The legally relevant question is not whether AI was used, how impressive its output appeared, or who owned the computer. It is whether each person proposed as an inventor actually conceived the subject matter claimed.

Author interpretation. The current guidance does not create a prompt-count threshold or declare that particular AI activities always qualify. Selecting a research problem, writing prompts, training a model, recognising a useful output, testing it, or modifying it may form part of the evidence. None of these labels settles inventorship by itself. Their significance depends on what the patent claims and what the person actually contributed to the conception of those claimed limitations.

Implications for AI-assisted research and development

The present rule does not prohibit the use of generative AI, optimisation systems or other computational models during invention. The USPTO compares AI systems with laboratory equipment, software and research databases: tools may provide services, ideas and assistance without becoming inventors.

That position produces two different outcomes:

1. A human conceives the claimed invention with AI assistance. AI use does not disqualify the human or create a separate heightened inventorship test. The application must name the correct natural person or people and meet every other condition of patentability.
2. No natural person conceived the claimed invention. Under the USPTO’s current interpretation, there is no qualifying human inventor to name. Naming the AI does not cure the problem because Thaler excludes non-natural persons from inventorship.

The second outcome follows from the agency’s application of the statute and conception doctrine. It should not be overstated as a broad judicial ruling on every autonomous-discovery system: Thaler did not decide an AI-assisted human claim, and no such factual record was before the court.

Inventorship, ownership and patentability are different

Inventorship identifies the natural person or people who conceived the claimed invention. Ownership asks who holds the resulting rights, which may depend on assignments, employment agreements and other law. Patentability tests the claimed subject matter against the Patent Act. These questions can lead to different answers.

For example, a company may own an application by assignment without being an inventor. A human may qualify as an inventor while the claim still fails because it is ineligible, anticipated, obvious or inadequately disclosed. Conversely, a technically novel AI output cannot receive a US patent merely because somebody owns the system that generated it.

A defensible development record

Author interpretation, not a USPTO checklist. Teams using AI in inventive work should preserve evidence that allows counsel and, if necessary, a court to reconstruct human conception. Depending on the project, that record may include:

• dated descriptions of the problem, constraints and proposed solution;
• versions of prompts, model outputs, code, drawings and experimental results;
• reasons for rejecting, selecting or materially changing an output;
• the human decisions that established each claimed feature or relationship;
• contributor records mapped to the eventual claims; and
• changes between an AI suggestion, a working embodiment and the subject matter finally claimed.

A prompt archive alone may show interaction without proving conception. A polished final specification may hide how the idea developed. Contemporaneous records are more useful when they identify who formed the particular solution and when, rather than merely documenting who operated the tool.

Policy considerations and unresolved questions

The original article moved too quickly from the existence of generative AI to claims about patent floods, market concentration, employment, disclosure incentives and economic growth. The primary legal authorities reviewed here do not establish those outcomes. They decide or administer a narrower question: who may be named as an inventor under the current Patent Act.

Observed legal position. In Thaler, the Federal Circuit rejected policy arguments that could not override unambiguous statutory text. If AI systems are ever to be recognised as inventors, Congress would need to confront the statutory definition and the connected rules for oaths, ownership, priority and enforcement. The court did not design that alternative regime.

Author interpretation. Before changing the law, policy analysis would need evidence on at least four matters:

• whether the absence of a human inventor causes socially useful inventions to be withheld, and how often;
• whether a new status for AI-generated inventions would improve disclosure or instead expand low-quality applications;
• who would receive initial rights and bear duties when a non-person generates the claimed subject matter; and
• how any new rule would interact with novelty, non-obviousness, enablement, priority and international filing practice.

Those are empirical and institutional questions. Assertions that recognising AI inventorship will necessarily accelerate innovation—or necessarily suppress it—remain scenarios unless supported by a defined dataset and causal analysis.

The immediate practical issue is more contained. Researchers should identify human conception honestly and at claim level. Patent applicants should not rely on the superseded 2024 “significant contribution” presentation as if it were still current USPTO guidance. Where the human contribution is uncertain, a registered US patent practitioner should assess the actual development record before filing or correcting inventorship.

Key primary authorities

• 35 U.S.C. § 100 — definitions of inventor and joint inventor
• 35 U.S.C. § 115 — inventor’s oath or declaration
• 35 U.S.C. § 116 — joint inventors and correction
• Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022) — precedential opinion
• Supreme Court docket 22-919 — certiorari denied 24 April 2023
• USPTO Revised Inventorship Guidance for AI-Assisted Inventions, 90 FR 54636 (28 November 2025)
• Superseded 2024 USPTO guidance, retained only to document the later correction