USPTO’s 2025 Revised Inventorship Guidance for AI-Assisted Inventions: What It Means for Inventors and Patent PractitionersDecember 3, 2025

In late 2025, the United States Patent and Trademark Office (USPTO) issued a Revised Inventorship Guidance for AI-Assisted Inventions (Docket No. PTO-P-2025-0014) that clarifies how inventorship is determined when artificial intelligence tools are involved. This new guidance explicitly rescinds the USPTO’s February 2024 inventorship guidance and replaces it with a framework grounded in traditional principles, reaffirming that only natural persons can be named as inventors on a patent application. Critically, the Office emphasizes there is no separate or modified legal standard for AI-assisted inventions: the familiar test of conception in the inventor’s mind remains the “touchstone of inventorship,” regardless of whether AI was used in the inventive process. This advisory blog post examines the 2025 guidance in the context of recent history and case law (including the now-rescinded 2024 guidance, Thaler v. Vidal, Pannu v. Iolab, and the DABUS saga) and analyzes the impact on patent prosecution and litigation strategy.

The question of AI inventorship moved from hypothetical to concrete with Dr. Stephen Thaler’s high-profile attempts to patent inventions generated by his AI system “DABUS” (Device for the Autonomous Bootstrapping of Unified Sentience). In 2019, Thaler filed U.S. patent applications naming DABUS as the sole inventor, openly stating that the creative machine, not any human, conceived the inventions (a new type of flashing beacon and a fractal food container design). The USPTO promptly issued a notice of missing parts, requiring a valid inventor’s name, and ultimately rejected the applications for failing to identify a natural-person inventor. Thaler’s petitions and appeals set off the first major judicial tests of whether an AI could be an inventor. By 2021–2022, the answer from the courts was unambiguous: only human beings can be inventors under current U.S. law. The Eastern District of Virginia and then the Federal Circuit in Thaler v. Vidal upheld the USPTO’s position, citing the Patent Act’s use of terms like “individual” (which the Supreme Court has said means a natural person absent contrary indication). The Federal Circuit found the statute “unambiguous in restricting inventors to natural persons” and concluded no further policy inquiry was needed. In April 2023, the U.S. Supreme Court declined to hear the case, letting this decision stand.

Importantly, Thaler v. Vidal left open a key question: what about inventions developed with significant AI assistance but involving human participation? The Federal Circuit explicitly noted it was not opining on whether a human using AI might lose entitlement to a patent if the AI’s contributions overshadow human contributions. This unresolved point, whether humans who “co-invent” with AI can still obtain valid patents, created uncertainty for AI-assisted R&D. Meanwhile, the DABUS episode had global ripple effects. Patent offices in Europe, the UK, and China similarly rejected DABUS applications for lack of a human inventor, reaching conclusions in line with the U.S. stance. One notable outlier was South Africa, which in 2021 granted a patent listing DABUS as inventor, effectively by administrative formality, since South Africa’s system did not substantively examine inventorship.

The USPTO and lawmakers were aware that AI’s role in innovation was growing and straining the inventorship framework. The Office took steps to study and get public input on these issues. By late 2022, even members of Congress became involved: Senators Thom Tillis and Chris Coons wrote to the USPTO and Copyright Office urging formation of a national commission to examine legal changes to better “promote…science and the useful arts” for AI-created works. In February 2023, the USPTO (Docket No. PTO-P-2022-0045) once again requested public comments on a host of AI and inventorship questions, signaling that policy evolution was being considered.

Against this backdrop, on February 13, 2024 the USPTO issued its first “Inventorship Guidance for AI-Assisted Inventions.” This 2024 guidance (now rescinded) acknowledged the rule from Thaler that an AI cannot be an inventor, but it ventured into how to analyze a human inventor’s contribution when AI is involved. It suggested that an inventor who uses an AI may qualify as such only if the person made a “significant contribution” to the conception of the invention and had recognition and appreciation of the invention’s features. Merely posing a problem to an AI or passively reviewing AI output would likely not be enough for inventorship, nor would simply owning or training the AI system by itself. The guidance gave examples: a person could demonstrate a significant contribution by skillfully crafting an AI prompt to yield a desired solution, or by materially tweaking and enhancing the AI’s output to arrive at the invention.

Notably, the USPTO explicitly imported the Pannu joint inventorship test into this context: requiring that each named inventor (in an AI-assisted project) satisfy the three Pannu factors for each claimed invention. If any one factor was not met for a given person, that person should not be listed as an inventor. The 2024 guidance warned that if evidence showed the human did not make a significant contribution to a claimed element (for instance, if the key inventive idea came entirely from the AI without human conceptual input), the application could face a rejection under 35 U.S.C. §§ 101 and 115 for lack of proper inventorship.

It also reminded applicants of their duty to disclose information material to patentability, which would include knowledge that the named inventor might not have actually conceived the invention, and a duty of reasonable inquiry to ensure inventorship is correct. In essence, the 2024 guidance took a cautious, stringent approach: using AI is permissible, but patent applicants had to be prepared to demonstrate that a human inventor truly conceived the invention (applying Pannu as a yardstick), claim by claim.

The 2024 guidance, however, was met with some criticism and confusion in the patent community. Many practitioners noted that the approach of applying Pannu factors, which normally govern multiple-human inventor scenarios, to a single-human-plus-AI situation was conceptually problematic. By definition, Pannu factors assume all putative “joint inventors” are persons; using them to effectively measure an AI’s contribution (a contribution that legally cannot qualify as inventorship at all) risked a logical paradox. Some feared that the 2024 guidance could lead to overzealous rejections or encourage examiners to second-guess inventorship in cases where an AI was merely a tool. It also raised practical questions: How would an examiner know the extent of AI involvement unless applicants self-disclosed it? Would applicants feel pressured to downplay an AI’s role to avoid a rejection, potentially at the cost of candor? These concerns, combined with continued stakeholder feedback, set the stage for the USPTO’s course correction in 2025.

On November 28, 2025, the USPTO’s new Revised Inventorship Guidance for AI-Assisted Inventions was published in the Federal Register, replacing the prior guidance in its entirety. The first notable change is an explicit rescission of the 2024 guidance’s approach of applying Pannu to AI situations. The USPTO acknowledges that the Pannu factors are only intended for determining whether multiple human persons are joint inventors, and that framework is “inapplicable when only one natural person is involved in developing an invention with AI assistance”. In other words, if an invention is developed by a single human working with the aid of AI, there is no “joint inventorship” question to analyze at all, because the AI by law cannot be a joint inventor.

The 2025 guidance makes clear that the same inventorship standard applies to all inventions, including those developed with AI assistance. There is no special, relaxed, or heightened test unique to AI-generated ideas. If an invention meets patentability requirements and a human is properly identified as the inventor, it does not matter that AI was used along the way. Conversely, if no human can be identified who meets the conception standard for the claimed invention, then patent protection is not available. In other words, the presence of AI does not change that outcome.

The guidance explicitly cites Federal Circuit precedent (Thaler v. Vidal) confirming that AI systems, no matter how sophisticated, cannot be inventors or co-inventors on a patent. Only natural persons can perform the legal act of invention. Thus, AI is treated as a tool, not an inventor. The Office drives this point home by analogizing AI systems to other kinds of aids inventors have historically used: “laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process”. Inventors and practitioners thus appear free to “use the services, ideas, and aid of others” (including AI) without those aides becoming co-inventors. This quote invokes principles as far back to O’Reilly v. Morse (1853) that an inventor can derive inspiration or information from external sources without losing their status as the inventor. In the AI context, this means that having a machine generate ideas or data is akin to having an extremely advanced research assistant or creative tool. It does not, in itself, jeopardize the human inventor’s claim, so long as the human truly conceived the invention as claimed.

Gregory Lars Gunnerson is a Partner, Intellectual Property Attorney in the Mechanical and Electrical Patent Practice Groups, and Chair of the Design Patent Practice Group at McKee, Voorhees & Sease, PLC. For additional information please visit www.ipmvs.com or contact Lars directly via email at gregory.gunnerson@ipmvs.com.