AI as Inventor: U.S. Patent Law and the Inventorship Question

The question of whether an artificial intelligence system can be named as an inventor on a U.S. patent application sits at the intersection of statutory text, administrative agency rulemaking, and federal appellate jurisprudence. This page examines the legal framework governing AI inventorship under U.S. patent law, including the USPTO's formal position, key court decisions, and the unresolved classification problems that arise when AI systems make non-trivial contributions to patentable discoveries. The stakes extend beyond individual patent filings to touch foundational questions about AI intellectual property law and who — or what — the patent system is designed to serve.

• Definition and scope
• Core mechanics or structure
• Causal relationships or drivers
• Classification boundaries
• Tradeoffs and tensions
• Common misconceptions
• Checklist or steps
• Reference table or matrix
• References

Definition and scope

Under 35 U.S.C. § 100(f), an "inventor" is defined as "the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention." The word "individual" has been the central fault line in every AI inventorship dispute litigated before U.S. courts. The United States Patent and Trademark Office (USPTO) has interpreted "individual" to mean a natural person — a human being — and has declined to accept applications that name an AI system as an inventor.

The scope of this question encompasses:

• Autonomous AI systems that generate novel outputs without explicit human direction at the claim level
• AI-assisted inventions where a human directs AI as a tool but the AI contributes conception
• Machine learning systems that surface patentable insights through training on large datasets
• Generative models that propose chemical compounds, mechanical configurations, or software architectures

The inventorship question is distinct from the question of patent eligibility (35 U.S.C. § 101) and patentability (35 U.S.C. §§ 102–103). An invention conceived entirely by an AI system may clear all patentability hurdles yet remain unpatentable solely because no qualifying human inventor can be named.

Core mechanics or structure

U.S. patent law imposes a duty of correct inventorship. Under 35 U.S.C. § 115, each inventor must submit an oath or declaration stating that the individual believes themselves to be an original inventor. An AI system cannot execute a legal oath.

Conception as the controlling standard. The Federal Circuit has long held that inventorship turns on conception — defined as "the complete performance of the mental part of the inventive act," specifically "the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention." This standard, articulated in Burroughs Wellcome Co. v. Barr Laboratories (Federal Circuit, 1994), presupposes a mind, and by extension a natural person. Courts applying this standard to AI systems have found no mechanism by which a machine can satisfy it.

The DABUS litigation. The most litigated AI inventorship case involves DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), an AI system developed by Dr. Stephen Thaler. Patent applications naming DABUS as sole inventor were filed in multiple jurisdictions. In the U.S., the USPTO rejected the applications. The Eastern District of Virginia upheld the USPTO's rejection in Thaler v. Vidal (2021), and the Federal Circuit affirmed in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), holding that the Patent Act "does not permit an AI to be an inventor." The Supreme Court declined to hear the case in 2023.

USPTO Inventorship Guidance for AI-Assisted Inventions (2024). In February 2024, the USPTO issued formal guidance (89 Fed. Reg. 10043) clarifying that AI-assisted inventions are not categorically unpatentable, but that every named inventor must be a natural person who made a "significant contribution" to the claimed invention, applying the Pannu factors from Pannu v. Iolab Corp. (Federal Circuit, 1998).

Causal relationships or drivers

Several intersecting forces have pushed AI inventorship to the forefront of patent policy.

Autonomous AI outputs in drug discovery. AI systems have identified novel molecular candidates in pharmaceutical research contexts where the design space is too vast for human exploration alone. When a system trained on protein-folding data proposes a structurally novel compound with measurable therapeutic activity, the human role may be limited to running the system and filing the resulting output — raising genuine uncertainty about who "conceived" the invention.

Commercial incentives and patent term duration. A granted U.S. utility patent provides 20 years of exclusivity from the filing date (35 U.S.C. § 154). Organizations investing heavily in AI-driven R&D have strong financial incentives to establish clear ownership of AI-generated innovations. Without a recognized AI inventorship pathway, those organizations face a choice: attribute inventorship to human collaborators whose contributions may not meet the legal conception threshold, or allow the innovation to fall into the public domain.

International divergence. Jurisdictions including South Africa (which granted a DABUS patent in 2021) and Australia (where a lower court initially accepted AI inventorship before being reversed on appeal) have followed different paths, creating pressure on U.S. policy. The World Intellectual Property Organization (WIPO) has tracked these divergences through its AI and IP Policy Initiative, noting inconsistent national treatment as of its 2022 report.

Legislative inaction. Congress has not amended 35 U.S.C. § 100 to address AI inventorship despite proposed legislative discussions. This vacuum leaves the USPTO and courts to interpret existing statutory text. The AI in U.S. legal system overview context illustrates how legislative gaps consistently push contested AI questions toward agency guidance and judicial interpretation.

Classification boundaries

Patent law recognizes three inventorship configurations relevant to AI-assisted work:

1. Human-only inventorship — AI used purely as a tool (comparable to a calculator or search engine); all conception attributable to identified human inventors. This is the cleanest category and faces no inventorship objections under current USPTO guidance.

2. AI-assisted human inventorship — Human inventors direct AI systems, evaluate outputs, select among alternatives, and apply the results to a defined problem. The 2024 USPTO guidance confirms this is patentable provided each named human inventor made a significant contribution to at least one claim under the Pannu factors: (a) contributed to the conception of the invention, (b) made a contribution that is not insignificant in quality, and (c) did more than merely explain the state of the art.

3. AI-autonomous output — The AI system generates the inventive concept without direction traceable to a specific human's mental act. Under current U.S. law, this category is unpatentable because no qualifying inventor can be named.

The boundary between categories 2 and 3 is actively contested and fact-specific. A researcher who formulates the problem, selects the training data, interprets the output, and refines the claim scope has a stronger inventorship argument than one who simply activates a commercial AI platform and files the first output it produces.

Tradeoffs and tensions

Disclosure incentive vs. inventorship barrier. The patent system's constitutional foundation (Article I, § 8, Clause 8) is the promotion of progress through the incentive of exclusive rights in exchange for public disclosure. If AI-autonomous inventions cannot be patented, the incentive to disclose evaporates. Developers may instead rely on AI trade secret law to protect AI-generated innovations, reducing the public knowledge base that patents are designed to build.

Accurate inventorship vs. practical workability. Requiring identification of the specific human mental act of conception in highly automated AI pipelines may force applicants into legally precarious inventorship attributions. Naming a human inventor whose contribution does not actually meet the Pannu threshold creates fraud risk under 35 U.S.C. § 256 and can render a patent unenforceable through inequitable conduct doctrine.

Competitive access vs. monopoly extension. Recognizing AI inventorship could enable a single organization with dominant AI infrastructure to accumulate patent portfolios at a scale impossible through human R&D, potentially distorting competitive markets in a way that runs counter to the patent system's balance. This tension intersects with AI antitrust law concerns that are separately developing.

Speed of innovation vs. statutory stability. AI R&D cycles now move faster than the legislative amendment process. The Federal Circuit's 2022 ruling in Thaler v. Vidal provides a clear rule but applies text written before modern machine learning existed. The clarity reduces litigation uncertainty while potentially misaligning incentives for the most advanced AI-driven industries.

Common misconceptions

Misconception 1: AI-assisted inventions are unpatentable in the U.S.
The USPTO's 2024 guidance explicitly states that AI-assisted inventions are not categorically unpatentable. The prohibition is narrow: AI systems cannot be named inventors. Inventions where a human made a significant contribution satisfying the Pannu factors remain patentable.

Misconception 2: The human who owns or operates the AI system is automatically the inventor.
Ownership and inventorship are legally distinct. Under 35 U.S.C. § 261, patents are assignable as property, meaning ownership can transfer. Inventorship, however, is not a matter of ownership or employment — it requires actual conception. An employer who owns an AI system that generates an invention is not thereby an inventor; the employer may receive the patent by assignment from a qualifying human inventor.

Misconception 3: Listing a human co-inventor cures any AI inventorship problem.
Naming a human co-inventor who did not actually contribute to the conception of at least one claim constitutes fraudulent inventorship. The duty of candor under 37 C.F.R. § 1.56 and the oath requirement under 35 U.S.C. § 115 make this a legally consequential choice, not a procedural workaround. A patent with incorrect inventorship can be held unenforceable.

Misconception 4: Other countries' recognition of AI inventors affects U.S. patent rights.
U.S. patent rights are territorial. A DABUS-derived patent granted in South Africa confers no U.S. patent rights and does not alter the USPTO's analysis of a corresponding U.S. application.

Misconception 5: The Thaler v. Vidal ruling resolved all AI inventorship questions.
The Federal Circuit's holding addressed only whether an AI system can be named as an inventor under the existing statute. It left unresolved the threshold question of how much human direction is sufficient to satisfy the Pannu factors when AI performs substantial portions of the inventive work.

Checklist or steps

The following sequence reflects the analytical steps involved in evaluating inventorship for an AI-assisted patent application under current USPTO framework. This is a descriptive process map, not legal advice.

Step 1 — Identify each claimed invention element.
Map each independent claim to the specific technical contribution it embodies. Claims that trace to purely human ideation are cleanest; claims that trace primarily to AI output require further analysis.

Step 2 — Trace the origin of each contribution.
For each claimed element, document the sequence of human decisions and AI-system outputs that produced it. Relevant factors include: who defined the problem; who selected or configured the AI system; who evaluated, modified, or rejected AI-generated alternatives.

Step 3 — Apply the Pannu factors to each proposed human inventor.
Under Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998), each named inventor must (a) contribute to the conception of the invention, (b) make a contribution not insignificant in quality, and (c) do more than explain the existing state of the art.

Step 4 — Assess whether any human satisfies the conception standard.
Apply the Burroughs Wellcome conception test: was there a definite and permanent idea, formed in a human mind, of the complete and operative invention?

Step 5 — Review the 2024 USPTO Guidance document.
Cross-reference the specific fact pattern against the USPTO's Inventorship Guidance for AI-Assisted Inventions (89 Fed. Reg. 10043) for illustrative examples and factor-weighing methodology.

Step 6 — Document the inventorship determination.
Maintain contemporaneous records of the AI system version used, training data sets, human-directed queries, and selection decisions. These records become relevant if inventorship is challenged post-grant.

Step 7 — Execute oaths and declarations only for qualifying inventors.
Under 35 U.S.C. § 115 and 37 C.F.R. § 1.63, only natural persons who actually believe themselves to be original inventors may execute the required declaration.

Step 8 — Consider alternative IP protections for any non-patentable AI-generated output.
If no human inventor satisfies the threshold, the innovation may qualify for trade secret protection, copyright protection of specific expression, or strategic publication to prevent competitor patenting (defensive publication).

Reference table or matrix

AI Inventorship Scenarios: Legal Status Under U.S. Law

International Treatment Comparison (Selected Jurisdictions)

The AI-generated content copyright parallel — where the U.S. Copyright Office similarly denies registration to purely AI-authored works — illustrates a consistent U.S. policy framework requiring human authorship or inventorship across both major IP regimes

References

• 35 U.S.C. § 100 – Definitions (Patent Act)
• 35 U.S.C. § 101 – Inventions Patentable
• 35 U.S.C. §§ 102–103 – Conditions for Patentability
• United States Patent and Trademark Office (USPTO)
• USPTO – Inventorship Guidance for AI-Assisted Inventions (Federal Register, Feb. 13, 2024)
• USPTO – Request for Comments on Artificial Intelligence and Intellectual Property Policy
• 37 C.F.R. § 1.41 – Applicant for Patent (eCFR)
• U.S. Copyright Office – Copyright and Artificial Intelligence (Policy Report, 2024)
• Federal Circuit Court Decisions – Thaler v. Vidal (Court of Appeals for the Federal Circuit)
• USPTO – Manual of Patent Examining Procedure (MPEP)