USPTO Replaces AI Inventorship Guidance

On November 28, 2025, the USPTO published new guidance on the Federal Register, replacing and rescinding its prior February 13, 2024 “Inventorship Guidance for AI-Assisted Inventions.” The new guidance makes clear that there will be no separate inventorship standard for inventions developed with AI assistance: the same legal standard that applies to traditional inventions applies to AI-assisted inventions. Consistent with long-standing case law, the USPTO reiterates that only natural persons—i.e., human beings—can be listed as inventors. AI systems, including generative AI or other computational models, are explicitly treated as tools comparable to lab equipment, software, or research databases.

Under the new framework, identification of inventors – whether a single human or multiple co-inventors – requires that at least one natural person satisfies the traditional standard of “conception” as defined under U.S. patent law. Conception remains “the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention, as it is hereafter to be applied in practice.” For single-inventor cases involving AI assistance, the inquiry becomes whether the human inventor’s mental conception (not just direction or oversight) produced the claimed invention. 

The new guidance also rejects the prior administration’s reliance on the Pannu v. Iolab Corp. three-factor test for “significant contribution” when only a single human inventor is involved. Under the 2025 guidance, the Pannu test applies only to joint inventorship scenarios involving multiple natural persons – because non-human AI cannot be a “joint inventor.”

The USPTO guidance also refers back to the seminal conception precedent of Burroughs Wellcome Co. v. Barr Labs., Inc to ground the analysis aimed at determining inventorship of such inventions when faced with challenges giving rise to such a question under 35 U.S.C. §§ 101 and 115. The guidance indicates that this analysis will remain “highly fact intensive” and revolve around the inventor’s ability to “describe [their] invention with particularity.”

That said, the traditional joint-inventor analysis remains relevant when two or more humans collaborate in an AI-assisted invention. In those cases, examiners and applicants should still consider whether each named inventor contributed in a “significant manner” to conception or reduction to practice, whether the contribution is “not insignificant in quality…relative to the entire invention,” and whether the contribution goes beyond merely explaining well-known aspects of the art.

Director John Squires directs examiners and judges to focus on classic conception principles, indicating that novelty will depend on how applicants and litigators contextualize the alleged inventor’s involvement in the formation of the inventive aspect of the claimed invention. For example, if a highly specific description of how an AI model arrives at the claimed aspect evidences conception, inventors and their employers would be incentivized to explain how their automated computing arrives at the ultimately identified solution. Conception would therefore rely on how well inventors and their employers contemporaneously grasp how their AI model works, which would necessitate companies to vigilantly supervise how the AI model adapts its own systems as well as how it arrives at certain outputs given the information fed to it. Ideally, Director Squire’s rescission will not allow for gamesmanship in front of the USPTO and district courts by encouraging applicants and litigators to claim contemporaneous understanding through a revisionist history of the invention story itself.

Some commenters have suggested that the new guidance may introduce uncertainty in the area of AI-assisted inventions.  Under the prior guidelines’ reliance on the Pannu factors, it could be said that, for AI-assisted inventions, so long as one human made a sufficient inventive contribution to be named as a co-inventor (if the inventive process were undertaken entirely by humans), then that person could be named as the inventor.  Under the new guidelines, which reiterate that only humans can be inventors, only the conception of the invention is to be considered.  The ongoing evolution of the inventorship standard for AI-assisted inventions also increases the likelihood that inventorship questions may be more heavily litigated than in the past.

From a practical standpoint, the 2025 guidance reinforces that AI-assisted inventions are not categorically unpatentable – they remain eligible for patent protection when properly conceived by human inventors. For patent practitioners and applicants, this means careful inventorship analysis remains essential in AI-assisted contexts, especially to ensure that the named inventors meet the traditional conception standard. Moreover, proper inventorship becomes even more critical in litigation or post-grant review settings. Companies using AI as part of their R&D processes should therefore pay increased attention to documenting human contributions to conception, particularly where AI plays a substantial role generating design variations, drafting code, identifying molecular candidates, or proposing engineering solutions. Organizations relying on AI tools should revise internal invention-disclosure procedures to require researchers to describe: (i) how AI was used; (ii) what prompts, models, or training data were involved; and (iii) how the human inventor shaped or refined AI-generated material into the final conception.