No. Under current U.S. law, an artificial intelligence system cannot be named as an inventor on a patent application, no matter how much of the inventive work it performed. This is settled by the Federal Circuit’s 2022 decision in Thaler v. Vidal and reaffirmed by the U.S. Patent and Trademark Office’s revised inventorship guidance, published in the Federal Register on November 28, 2025. What AI CAN do is assist a human inventor — and inventions developed with heavy AI assistance remain fully patentable, as long as at least one natural person made a significant contribution to conceiving the claimed invention. Confusing those two questions is the most common and most consequential mistake research administrators and inventors make when AI tools show up in a lab notebook. This guide separates them, walks through the legal history that produced the current rule, and sets out what technology transfer offices (TTOs) should actually do when a disclosed invention involved AI assistance.
The short answer
- AI cannot be an inventor. Only a natural person (a human being) can be listed as an inventor or joint inventor on a U.S. patent, under 35 U.S.C. § 100(f) as construed by the Federal Circuit in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022).
- AI-assisted inventions remain patentable. The use of an AI tool anywhere in the inventive process does not disqualify an invention from patent protection, provided a natural person contributed significantly to conceiving the claimed invention.
- The USPTO no longer applies a separate legal test to AI-assisted inventions. Its November 2025 guidance rescinds the February 2024 guidance and eliminates the AI-specific version of the Pannu joint-inventorship factors that the 2024 guidance had introduced. The same conception-based inventorship analysis the USPTO has always applied now governs AI-assisted inventions too.
- For TTOs, the practical burden hasn’t gone away — it has shifted. Institutions still need documentation showing a human inventor’s contribution to conception, but the standard for what counts is now the ordinary inventorship standard, not a bespoke AI checklist.
Thaler v. Vidal: why AI cannot be named an inventor
The legal question of whether an AI system can be a named inventor was tested directly by computer scientist Stephen Thaler, who filed two patent applications identifying an AI system he built, called DABUS (“Device for the Autonomous Bootstrapping of Unified Science”), as the sole inventor of a food container and a light beacon. The USPTO rejected both applications for failing to name a valid inventor. Thaler sought judicial review, and the U.S. District Court for the Eastern District of Virginia granted summary judgment for the USPTO.
On appeal, the Federal Circuit affirmed in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022). The court’s reasoning was a straightforward statutory-text analysis: the Patent Act defines “inventor” (35 U.S.C. § 100(f)) as “the individual” who invents, and elsewhere consistently uses personal pronouns (“himself,” “herself,” “individual”) to describe inventors. The court held that this language limits inventorship to natural persons, consistent with earlier Federal Circuit precedent holding that neither corporations nor states/sovereigns can be inventors for the same textual reason. The court did not reach the broader policy question of whether AI-generated inventions should be patentable at all — only that, as written, the statute requires a human inventor.
Thaler petitioned the U.S. Supreme Court for review. The Supreme Court denied certiorari on April 24, 2023, leaving the Federal Circuit’s holding as the governing law. (Thaler pursued a parallel line of litigation on the copyright side — over whether an AI-generated work can be copyrighted without a human author — which the Supreme Court also declined to take up on further appeal.) For patent purposes, Thaler v. Vidal is the controlling precedent: an inventor must be a natural person, full stop.
The 2024 guidance, and why the USPTO rescinded it
Following Thaler, the USPTO still had to answer a narrower, more common question than DABUS’s: not “can an AI be an inventor” (settled: no), but “does an invention that a human developed with substantial help from an AI tool still qualify for a patent, and if so, what does the human need to have contributed?” On February 13, 2024, the USPTO published its first formal answer: the “Inventorship Guidance for AI-Assisted Inventions.” That guidance confirmed AI-assisted inventions can be patented, but built its analysis around the Pannu factors — a three-part test (Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998)) originally developed to decide whether someone qualifies as a joint inventor alongside other named human inventors — and adapted that framework specifically to evaluate a human’s contribution relative to an AI system’s output.
That adaptation drew sustained criticism from patent practitioners over the following year and a half: the Pannu factors were designed to compare contributions between two or more people, not to police the boundary between a human contributor and a non-human tool, and applying them to AI assistance produced an analysis that was, in practice, harder to satisfy and less predictable than the inventorship standard applied everywhere else in patent law.
The revised November 2025 guidance: what actually changed
The USPTO published “Revised Inventorship Guidance for AI-Assisted Inventions” in the Federal Register on November 28, 2025 (90 FR 54636, Docket No. PTO-P-2025-0014, Federal Register document number 2025-21457). The revised guidance rescinds the February 13, 2024 guidance in its entirety and replaces it.
The substantive change is a simplification, not a loosening or tightening of the underlying rule: the revised guidance eliminates the AI-specific application of the Pannu factors and states that the same legal standard for determining inventorship applies to every invention, regardless of whether AI systems were used anywhere in the inventive process. Under this standard, an AI system — whether generative AI, another machine-learning model, or any other computational tool — is treated the same way the law has always treated a sophisticated piece of lab equipment, analytical software, or a research database: as a tool a human inventor may use, not as a contributor whose output has to be legally disentangled from the human’s own conception.
Practically, this means the Pannu factors are back to their original, narrower role: they apply when determining whether two or more natural persons each qualify as a joint inventor, and they simply do not come into play in a single-inventor case where that person used AI tools during the work. The relevant question reverts to ordinary conception-of-invention doctrine that has always governed inventorship: did a natural person form a definite and permanent idea of the complete and operative invention? If yes, AI assistance in getting there — generating candidate compounds, running simulations, suggesting design variations, drafting analysis — does not change who the inventor is. The guidance applies across utility, design, and plant patents.
What “significant contribution” means now
Because the revised guidance folds AI-assisted inventorship back into ordinary conception doctrine, there is no separate AI checklist to apply. The question a TTO or patent counsel needs answered is the same one asked of any claimed inventor: did this person conceive the invention — form the definite, permanent idea of the complete, operative invention, including how to reduce it to practice — or did they merely carry out someone (or something) else’s already-formed idea?
Under that standard, using an AI tool the way a researcher uses any other instrument — to run a search, screen candidates, generate a first-pass draft, model an outcome, or accelerate routine calculation — does not diminish the human’s claim to inventorship, in the same way that operating a mass spectrometer or a statistical package has never diminished a researcher’s claim to inventing whatever the equipment helped them discover. What still would not qualify a person as an inventor is the same thing that has never qualified anyone as an inventor: merely posing a problem to be solved, merely recognizing a result’s significance after the fact, or merely reducing someone else’s already-conceived idea to practice, without contributing to the conception itself. The difference an AI tool introduces is factual, not legal — institutions still have to be able to show, with real contemporaneous evidence, which human made which conceptual contribution when several people (and one or more AI tools) were all involved in a project.
Practical guidance for TTOs and research administrators
The legal question is now simpler than it was under the 2024 guidance, but the evidentiary burden on institutions has not gone away — if anything, it is more important to get documentation right now that there is no separate AI-specific safe harbor to lean on. A few practices follow directly from the revised guidance:
1. Update invention disclosure intake to ask about AI tool use directly
A disclosure form should ask, plainly, whether any AI tool was used in the course of developing the claimed invention, and if so, how (literature/prior-art search, hypothesis generation, molecule/design screening, data analysis, drafting). This is not because AI use changes whether the invention is patentable — it doesn’t, on its own — but because it flags cases where the office needs to look more carefully at exactly who contributed the conception, especially on multi-person projects where an AI tool’s output might otherwise get informally attributed to “the team” rather than to a specific individual.
2. Document conception with contemporaneous, individual-level records
Lab notebooks, electronic lab notebook entries, dated design documents, and version-controlled model/code repositories should make clear which named individual formed which idea, and when, independent of what any AI tool output. This is the same evidentiary discipline that has always mattered for joint inventorship determinations under the Pannu factors — it now matters even more because it is the only test being applied; there is no separate, more lenient AI-specific standard to fall back on if the human-conception record is thin.
3. Train inventors and lab staff on the distinction before disclosure, not after
Researchers who assume “the AI came up with it” often mean something more specific and more favorable to patentability than they realize — usually that the AI tool generated candidates or output that a human then evaluated, selected, and directed toward a specific, useful application. That evaluation-and-direction step is frequently where legally significant conception actually happens. TTO staff who interview inventors at disclosure time should ask concretely: who decided which AI-generated output was worth pursuing, who modified it, and who recognized why it would work, rather than accepting “AI generated this” as a reason to assume the invention can’t be protected.
4. Don’t let AI-assistance questions get conflated with inventorship-correction questions
If an AI-assistance issue surfaces only after a patent has already issued — for instance, a dispute later emerges over whether a named inventor’s contribution was actually AI output they merely reviewed — that is a question about correcting inventorship on an issued patent under 35 U.S.C. § 256, a different (and generally more expensive) procedural track than getting the disclosure and initial filing right in the first place.
5. Keep AI-assistance disclosure separate from federal-funding reporting obligations
None of the above changes an institution’s underlying Bayh-Dole invention-disclosure and reporting obligations for federally funded inventions; AI-tool use is a fact to document as part of describing how the invention was made, not a separate compliance regime layered on top of existing iEdison and election-of-title timelines.
Common mistakes to avoid
- Treating “AI helped make this” as a reason to reject a disclosure outright. The correct question is whether a human contributed significantly to conception, not whether AI was involved at all — almost every modern lab uses AI-enabled software somewhere in its workflow.
- Assuming the February 2024 guidance’s AI-specific Pannu-factors analysis still applies. It was rescinded in full by the November 28, 2025 revised guidance; citing the 2024 framework in an office action response or an internal TTO policy is now citing withdrawn guidance.
- Confusing “AI can’t be an inventor” with “AI-assisted inventions can’t be patented.” These are not the same rule. Thaler v. Vidal answers only the first question. AI-assisted inventions with a genuine human contributor to conception are patentable exactly like any other invention.
- Skipping documentation because “the legal test is simpler now.” A simpler legal standard does not reduce the practical need for contemporaneous, individual-attributed conception records — it just means those records are being measured against the same bar as every other inventorship determination, not a separate and possibly more forgiving AI-specific one.
Frequently asked questions
Can an AI system be listed as an inventor on a U.S. patent application?
No. Under Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), and the USPTO’s implementing guidance, only a natural person can be named as an inventor. This has not changed with the November 2025 revised guidance — it is the settled premise the revised guidance builds on, not something it reconsiders.
Does using AI tools during research make an invention unpatentable?
No. An invention developed with AI assistance is patentable on the same terms as any other invention, provided a natural person made a significant contribution to conceiving the claimed invention. The USPTO’s revised guidance treats AI tools as tools, comparable to lab equipment or analytical software, not as something that taints an invention’s eligibility.
What happened to the Pannu factors for AI-assisted inventions?
The February 13, 2024 USPTO guidance had adapted the Pannu joint-inventorship factors (from Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998)) into a bespoke test for evaluating a human’s contribution against an AI system’s output. The November 28, 2025 revised guidance withdraws that AI-specific application. The Pannu factors still exist and still apply, but only in their original role: deciding whether multiple named human inventors each qualify as joint inventors.
Did the Supreme Court ever rule on AI inventorship?
The Supreme Court denied certiorari in Thaler’s petition on April 24, 2023, declining to review the Federal Circuit’s decision. That left Thaler v. Vidal as the final word from the courts; the Supreme Court has not itself ruled on the merits of AI patent inventorship.
What should a university require on its invention disclosure form regarding AI tools?
At minimum, whether any AI tool was used in developing the invention and, if so, in what capacity (search, screening, generation, analysis, drafting). The purpose is not to disqualify AI-assisted work but to make sure the disclosure record clearly shows which named human contributed the conception, which is the actual legal question the USPTO and, eventually, a court would ask if inventorship were ever challenged.
This guide reflects publicly available primary sources as of July 2026: the Federal Circuit’s opinion in Thaler v. Vidal and the USPTO’s Federal Register notices for its February 2024 and November 2025 inventorship guidance. It is provided for research-administration reference and is not legal advice; institutions should confirm current guidance directly with USPTO publications and with patent counsel before making inventorship determinations on specific disclosures.
