Human Conception in AI-Assisted Invention and Patent Drafting: Recognition and Original Inventorship After Thaler and the Revised USPTO Guidance
Executive Summary: This article argues that AI’s generation of claim language or technical content does not itself resolve human inventorship. Drawing on traditional conception doctrine, Thaler v. Vidal, and the USPTO’s November 2025 revised guidance, it distinguishes AI’s articulation of an existing human invention, a human’s completion of an AI-suggested solution, and a human’s comprehension of a complete machine-produced design. It advances the position that sufficiently definite technical comprehension of an unforeseen AI output can constitute original human conception, while acknowledging that this application remains unsettled and requires factual and doctrinal support. Mere prompting, selection, approval, or commercial appreciation does not establish inventorship. The analysis also preserves ordinary joint-inventorship principles, including the need to assess substantive technical contributions by patent counsel, and explains why later human conception cannot cure inadequate disclosure in an application as filed. For practitioners, it recommends identifying the operative technical idea, documenting the relevant human understanding and contributions, and evaluating disclosure support separately. The proposed approach accommodates substantial AI assistance while retaining original human conception as the basis for inventorship and recognizing that innovation and disclosure incentives cannot substitute for the legal requirements.
I Introduction
Generative AI increasingly enters patent practice while inventors and counsel are translating technical disclosures into claims. A model may supply clearer language for an existing idea, suggest an incomplete alternative, or produce a detailed solution that no participant previously considered. Saying that AI “originated” the resulting limitation does not distinguish these situations. Patent law asks who conceived the claimed invention, and identifying the first source of a sentence does not necessarily answer that question. Burroughs Wellcome Co. v. Barr Lab’ys, Inc., 40 F.3d 1223, 1227–28 (Fed. Cir. 1994); Dana-Farber Cancer Inst., Inc. v. Ono Pharm. Co., 964 F.3d 1365, 1371–72 (Fed. Cir. 2020).
Two propositions frame the inquiry. Under Thaler v. Vidal, only natural persons can be inventors under the Patent Act. Under the USPTO’s November 2025 revised guidance, the traditional conception standard applies whether or not AI assisted the inventive process. Neither proposition, standing alone, establishes that a human who understands and approves AI output has conceived the invention. Thaler expressly left the patentability of inventions made by humans with AI assistance outside its decision, and the guidance retains a fact-intensive inquiry into human conception. Thaler v. Vidal, 43 F.4th 1207, 1212–13 (Fed. Cir. 2022); Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. 54,636, 54,636–37 (Nov. 28, 2025).
This article argues that AI’s generation of technical content should not itself defeat human inventorship. Human conception may precede the output, develop through interaction with it, or arise through a sufficiently definite understanding of an unforeseen result. The last possibility presents the difficult case. Recognition can describe anything from noticing that an output looks promising to comprehending a complete operative invention. A persuasive conception analysis must distinguish those activities rather than treat prompting, selection, and approval as a checklist that automatically establishes inventorship.
The analysis builds on scholarship examining conception and causation, the difficulties of allocating inventorship within collaborative research, and the limitations of treating AI as an isolated autonomous actor. The proposed account preserves the requirement of original human conception while rejecting an automatic inference from machine-generated content to the absence of a human inventor. See Dan L. Burk, Causation and Conception in American Inventorship, 20 Duke L. & Tech. Rev. 116, 122–28 (2023); Amy R. Motomura, The Inventorship Fallacy, 58 U.C. Davis L. Rev. 2379, 2381–84 (2025); Liza Vertinsky, Boundary-Spanning Collaboration and the Limits of Joint Inventorship Doctrine, 55 Hous. L. Rev. 401, 443–48 (2017); James Bryan Williams, AI and Inventorship: An Engineering Perspective on the Current Debate (rev. Nov. 14, 2024) (unpublished manuscript), https://ssrn.com/abstract=4064320.
II The Governing Conception Standard
Section 100(f) defines an inventor by reference to the individual or individuals who invented or discovered the subject matter. Section 115(b)(2) requires the inventor’s oath or declaration to state the individual’s belief that he or she is the original inventor or an original joint inventor of a claimed invention. Conception, in turn, requires a definite and permanent idea of the complete operative invention. A general objective or research plan is insufficient; the idea must be settled enough that a person of ordinary skill could reduce it to practice without extensive research or experimentation. 35 U.S.C. §§ 100(f), 115(b)(2); Burroughs Wellcome, 40 F.3d at 1227–28; Sewall v. Walters, 21 F.3d 411, 415 (Fed. Cir. 1994).
The conception requirement concerns the technical invention, not the particular vocabulary chosen to describe it. An inventor may possess an operative arrangement before knowing how to express it in claim language. Conversely, a fluent claim can describe subject matter that no named human has conceived. The relevant evidence must connect human understanding and contribution to the invention defined by the claims. Neither polished drafting nor the decision to seek protection substitutes for that connection. See Solomon v. Kimberly-Clark Corp., 216 F.3d 1372, 1381–82 (Fed. Cir. 2000); Sewall, 21 F.3d at 415.
Conception does not invariably require experimental confirmation. Dana-Farber distinguishes formation of the inventive idea from later verification and recognizes that significant contributions can emerge across an extended collaboration. That distinction prevents the final experiment from eclipsing earlier conception. It does not make every useful preliminary activity inventive, or eliminate the need to identify what each participant contributed to the claimed subject matter. Dana-Farber, 964 F.3d at 1371–73.
The revised guidance rescinds the February 2024 guidance in its entirety and withdraws its application of the Pannu factors to a single human working with AI. The stated reason is straightforward: an AI system is not a person, so that situation presents no joint-inventorship question. When multiple humans participate, however, traditional joint-inventorship principles, including Pannu, still determine which humans qualify. The revision does not exempt a project from those principles merely because a model participated. Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,636–37; Pannu v. Iolab Corp., 155 F.3d 1344, 1351 (Fed. Cir. 1998).
The guidance characterizes AI as an instrument comparable to software, laboratory equipment, and research databases. Its supporting authorities recognize that inventors can obtain information and assistance from others without making every source a joint inventor. That analogy rejects machine personhood and permits substantial assistance; it does not dispense with the antecedent requirement that a human conceive the invention. The notice states the USPTO’s examination position. The applications of that position developed below are arguments about conception, not holdings supplied by the notice. Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,637 & n.16 (citing Shatterproof Glass Corp. v. Libbey-Owens Ford Co., 758 F.2d 613, 624 (Fed. Cir. 1985), and Hess v. Advanced Cardiovascular Sys., Inc., 106 F.3d 976, 981 (Fed. Cir. 1997)).
III Recognition and Original Human Conception
The most difficult issue is whether a human can conceive an invention by comprehending a complete solution first produced by AI. It is tempting to answer by identifying who generated the content. But a machine’s causal role in producing an output and a person’s legal status as an inventor are distinct questions. Thaler resolves the machine’s eligibility for that status. It does not establish that the remaining human must qualify, or determine which acts of recognition satisfy conception. Thaler, 43 F.4th at 1212–13; see Dan L. Burk, Causation and Conception in American Inventorship, 20 Duke L. & Tech. Rev. 116, 122–28 (2023).
Burk supplies an affirmative account of recognition that deserves direct engagement. He separates conception from the material work that produces an embodiment and explains that the temporal order need not run from conception to construction. His hypothetical includes an automated result that the human did not previously envision: the human may become an inventor upon comprehending the invention. This supports the possibility that an unforeseen machine output can become the subject of human conception after it exists. It is a scholarly account of the doctrine, however, rather than an appellate holding about a person adopting a complete generative-AI proposal. Dan L. Burk, Causation and Conception in American Inventorship, 20 Duke L. & Tech. Rev. 116, 122–25 (2023).
The distinction between recognizing a promising result and conceiving an operative invention is essential to that account. Recognizing that a proposed controller might reduce energy use does not establish possession of its operative architecture. Understanding the components, their relationships, and the conditions under which the claimed process operates presents a stronger case. The proposed approach asks whether the human’s engagement amounts to formation of the complete technical idea, rather than assuming that any expression of approval supplies the missing conception. It does not require a human to alter a machine’s proposal merely to demonstrate independence.
Nor does the proposal make detailed understanding universally sufficient. Burk’s derivation discussion explains that someone who receives another human’s completed invention may understand it fully without acquiring the status of its original inventor. That recipient has a mental picture but lacks an original inventive conception. A recognition-based account of AI assistance must therefore explain why receiving machine output differs from appropriating a human conceiver’s contribution. The absence of a legally eligible machine inventor makes the situations different, but it does not, by logical necessity, establish the recipient’s entitlement. Dan L. Burk, Causation and Conception in American Inventorship, 20 Duke L. & Tech. Rev. 116, 126–28 (2023).
The argument advanced here is that technical comprehension of an unforeseen machine result can constitute original human conception when it is the act through which the person first forms the definite operative idea, rather than acquires another human’s completed conception. The argument is strongest where contemporaneous evidence shows how the person understood the operative combination and distinguished it from incomplete or unsuitable alternatives. It is weaker where the record shows only acceptance of a finished answer. The quantity of prompting or effort does not resolve that distinction. These considerations identify relevant evidence; they are not a new AI-specific legal test.
In re VerHoef supplies an important limiting comparison. The applicant identified a problem, pursued its solution, and claimed intellectual control, but another human supplied the essential figure-eight feature of the claimed dog harness. Applying pre-AIA § 102(f), the court rejected sole inventorship. A person’s direction of the project did not absorb another person’s contribution. Because VerHoef involved human collaborators, it does not settle the machine-output question. It nevertheless defeats any general proposition that framing a problem, supervising development, and choosing a result necessarily establish sole conception. In re VerHoef, 888 F.3d 1362, 1365–67 (Fed. Cir. 2018).
IV Three Technical Scenarios
Consider a cooling controller that receives temperature measurements and switches a fan between operating states. The following scenarios hold that setting constant while changing the technical work performed by the human and the model. Each assumes that the claims ultimately identify the operative control arrangement, rather than merely a desired result.
In the first scenario, the inventor’s disclosure already specifies separate activation and deactivation thresholds and the state transitions between them. Counsel asks AI to express the arrangement as a method claim. The model supplies the term “hysteresis” and reorganizes the steps without adding a technical relationship. The model has contributed wording for an already-conceived solution. Inventorship does not turn on whether the inventor had previously used that word. The comparison between the original disclosure and the proposed claim supplies the relevant evidence of prior human conception. See Solomon, 216 F.3d at 1381–82; Sewall, 21 F.3d at 415.
In the second scenario, the inventor initially discloses only threshold-based switching. The model suggests adapting the thresholds to changing conditions but leaves the mechanism unspecified. The inventor determines which measurements govern adaptation, derives a rule for updating the thresholds, and adds a state transition that prevents unstable switching. Those acts concern the technical solution itself. If they yield a definite operative combination reflected in the claim, they provide a substantial basis for human conception. The analysis turns on those contributions and the resulting invention, not on the mere fact that the inventor selected a model-generated suggestion. See Burroughs Wellcome, 40 F.3d at 1227–28; Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,637.
In the third scenario, the model supplies the complete adaptive controller, including the update rule, state transitions, and parameter relationships. The human studies the proposal, understands its operation, and elects to claim it without alteration. This is the recognition problem in its strongest form. The first scenario’s prior conception and the second scenario’s completion of an incomplete design cannot simply be assumed. Treating recognition as conception in this setting depends on the reasoning developed above. The absence of technical alteration does not automatically defeat that argument, but comprehension and approval cannot be treated as an established safe harbor. See Dan L. Burk, Causation and Conception in American Inventorship, 20 Duke L. & Tech. Rev. 116, 124–28 (2023); Thaler, 43 F.4th at 1213.
The facts matter within the third scenario. A record showing only that someone clicked “accept” does not demonstrate a definite understanding of the invention. A contemporaneous explanation of the control rule, its necessary relationships, and how it solves the identified problem is stronger evidence. Experimental work may illuminate that understanding or reveal further inventive development; routine testing alone does not resolve who conceived the solution. The point is to identify the human mental act and its technical content rather than equate the existence of a working output with human inventorship. See Dana-Farber, 964 F.3d at 1372–73; Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,637.
A fourth variation exposes the importance of human attribution. If counsel, rather than the client, develops the missing adaptive rule in the second scenario, calling the result “human conceived” does not complete the analysis. The article’s proposed treatment of machine output leaves intact the separate inquiry into which natural persons contributed to conception. A claim that changes during development may require a corresponding reassessment of the humans who qualify as inventors. 35 U.S.C. § 116(a); Pannu, 155 F.3d at 1351; Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,637.
V Counsel and Other Human Contributors
Patent counsel ordinarily helps the client describe an invention, distinguish claim scope, and select a form of protection. Those activities should not be aggregated with technical development under an undifferentiated reference to “the inventor and attorney.” Solomon rejected an inventorship challenge that relied on imprecision in the inventor’s testimony and an assertion that counsel might be the true inventor. The court emphasized counsel’s proper role in defining the client’s invention. That reasoning supports the distinction between legal formulation and invention, not a categorical exemption for every technical contribution an attorney might make. Solomon, 216 F.3d at 1381–82.
When counsel supplies a technical relationship that contributes to conception, the ordinary standards for joint inventorship remain relevant. The same is true of engineers, researchers, and other participants who prompt or evaluate a model. Each asserted contribution must be connected to the claimed invention, and the humans must satisfy the required collaborative relationship. Kimberly-Clark and CODA help define that relationship among people; there is little need to repurpose their collaboration requirement as an independent theory for excluding AI, which is already excluded by the natural-person requirement. Kimberly-Clark Corp. v. Procter & Gamble Distrib. Co., 973 F.2d 911, 917 (Fed. Cir. 1992); CODA Dev. s.r.o. v. Goodyear Tire & Rubber Co., 916 F.3d 1350, 1358–61 (Fed. Cir. 2019).
Falana illustrates why labels such as “assistance” are insufficient. The contributor’s development of a synthesis protocol supported joint inventorship because it contributed to conception of the claimed compounds. His contribution was not discounted simply because it supplied a method used by others. Conversely, general instruction or routine implementation does not necessarily establish inventorship. Applied to the controller examples, the inquiry is whether a person contributed to the operative claimed solution, rather than whether that person occupied a supporting role or worked through an AI interface. Falana v. Kent State Univ., 669 F.3d 1349, 1357–59 (Fed. Cir. 2012).
A claim-centered inquiry also need not become a search for the person who supplied the single feature that secured allowance. Dana-Farber examines the significance of contributions within the collaborative development of the claimed invention and distinguishes that inquiry from novelty over a participant’s earlier work. At the same time, criticism of excessive focus on individual limitations does not justify disregarding the claims. The claims identify the subject matter for which inventorship must be established; evaluating the combination supplies context for assessing a contribution’s significance. Dana-Farber, 964 F.3d at 1371–73; see Liza Vertinsky, Boundary-Spanning Collaboration and the Limits of Joint Inventorship Doctrine, 55 Hous. L. Rev. 401, 445–48 (2017).
VI The Original Inventor Requirement
The phrase “original inventor” in § 115 has substantive work to do. The statutory declaration concerns the individual’s belief in original inventorship; it is not a declaration that the individual personally typed every claim phrase. Nevertheless, replacing a first-expression inquiry with conception doctrine must not make originality disappear. A person cannot establish original conception merely by understanding and restating another human’s completed invention. 35 U.S.C. § 115(b)(2); see Dan L. Burk, Causation and Conception in American Inventorship, 20 Duke L. & Tech. Rev. 116, 126–28 (2023).
The distinction relevant here is between the origin of language and the origin of the claimed technical idea. An AI system that translates a detailed disclosure into a claim does not displace the human’s earlier conception. When AI supplies previously unconsidered technical substance, however, the inquiry cannot stop at linguistic authorship. It must determine whether the subsequent human activity constitutes original conception under the defended recognition account, or instead leaves the project without a qualifying human inventor for the claim as drafted.
The machine’s legal incapacity to invent does not require patent law to award the resulting subject matter to someone else. That possible gap marks a substantive limit of the proposed account. A claim may lack a qualifying human inventor even though the output is useful and understandable. Equally, the possibility of that gap does not establish that every unforeseen machine suggestion prevents human conception. The proper task is to identify and justify the circumstances in which the human’s technical understanding amounts to the originating conception required by law. See Thaler, 43 F.4th at 1212–13; Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,636–37.
Care is also necessary when invoking commentary on the USPTO’s evolving position. Aboy, Liddell, and Lath’s 2025 article examined the February 2024 guidance, including its significant-contribution framework and examples. It remains useful for understanding that earlier approach and the possibility of inventorless inventions, but cannot establish the meaning of the later revision. Aboy and Liddell’s 2026 follow-up addresses the November 2025 notice directly, including withdrawal of the Pannu analysis for single-human AI assistance. The two accounts must be situated in their respective administrative settings. Mateo Aboy, Kathleen Liddell & Aparajita Lath, Inventorship in the Age of AI: Examining the USPTO Guidance on AI-Assisted Inventions, 20 J. Intell. Prop. L. & Prac. 495, 495–502 (2025); Mateo Aboy & Kathleen Liddell, Revised USPTO Guidance on Inventorship for AI-Assisted Inventions: A Pro-Innovation Pivot Away from Pannu Factors, 21 J. Intell. Prop. L. & Prac. 272, 272–75 (2026), https://doi.org/10.1093/jiplp/jpag021.
VII Conception Evidence and the Filing Boundary
The three scenarios suggest a practical way to organize evidence without creating a special legal test for AI. The relevant record identifies the technical disclosure before model use, the substance introduced by the output, and the human’s ensuing explanation or development of the operative solution. Prompts and output histories can help establish chronology, but their volume proves little by itself. A concise technical account of what the human understood or supplied is more probative than a generalized statement that the human directed the process. See Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. at 54,637 (describing the factual inquiry and the inventor’s ability to describe the invention with particularity).
This evidentiary recommendation concerns substantiating conception. It does not make prompt logging a statutory prerequisite or displace the USPTO’s general presumption that named inventors are the actual inventors. Nor should the proposed recognition analysis be confused with measuring human effort against the machine’s effort. The inquiry concerns the claimed technical idea and the legally relevant human acts, not an allocation of percentages of creative work. Id. at 54,636–37.
Timing introduces an independent limitation in prosecution. Before an application is filed, a newly developed embodiment can be assessed for inventorship and included in the application with adequate disclosure. After filing, a claim amendment must find support in the original disclosure. Section 112(a) governs written description and enablement, while § 132(a) prohibits amendments introducing new matter into the disclosure. A human’s later conception of an AI-suggested feature does not retrospectively supply missing original support. 35 U.S.C. §§ 112(a), 132(a); Manual of Patent Examining Procedure §§ 2163(I)(B), 2163.06 (9th ed., rev. 01.2024, Nov. 2024).
For the cooling controller, a post-filing amendment that restates an already-disclosed pair of switching thresholds differs from one that adds an undisclosed adaptive update rule. The first may present a drafting question; the second raises a support problem even if a human unquestionably conceived the new rule. A later application may provide a vehicle for new subject matter, but that subject matter does not acquire the earlier filing date merely because it emerged while prosecuting the earlier case. Inventorship and disclosure must therefore be analyzed separately. See 35 U.S.C. § 120; Manual of Patent Examining Procedure § 2163(II)(A)(3)(b) (9th ed., rev. 01.2024, Nov. 2024).
VIII Patent Policy and the Competing Incentives
Policy supports taking recognition seriously, but does not establish that every technically useful AI output deserves a patent. A rule treating substantial AI involvement as inherently disqualifying could discourage useful tools or encourage firms to favor secrecy. Those are plausible incentive effects, not demonstrated consequences of every conception boundary. The existing literature identifies concerns about unprotected AI-produced inventions and disclosure, but further evidence is needed to establish how different rules would affect particular industries. See David L. Schwartz & Max Rogers, “Inventorless” Inventions? The Constitutional Conundrum of AI-Produced Inventions, 35 Harv. J.L. & Tech. 531, 564 & n.193 (2022); Mateo Aboy, Kathleen Liddell & Aparajita Lath, Inventorship in the Age of AI: Examining the USPTO Guidance on AI-Assisted Inventions, 20 J. Intell. Prop. L. & Prac. 495, 499–502 (2025).
The strongest objection to a broad recognition account is that comprehension may become an inexpensive route to exclusive rights over machine-produced solutions. If automation substantially lowers the cost of generating candidate inventions, the incentive needed to produce those candidates may also fall. A permissive rule could increase the number of patents and the cost of navigating them without a corresponding increase in socially valuable research. These are competing predictions requiring analysis, rather than reasons to assume either that AI-assisted products are superior or that patent exclusion necessarily causes a market failure.
A sound policy analysis must consider implementation, testing, commercialization, and disclosure, not merely candidate generation. Those activities do not themselves establish inventorship. The recognition account addresses who conceives; its policy justification asks whether treating qualifying recognition as conception fits the patent system’s incentives. Favorable consequences cannot supply a missing doctrinal element.
Thaler rejected policy appeals as grounds to override the natural-person requirement and explained that § 103’s treatment of the manner of invention does not resolve who may be an inventor. Policy can inform conception doctrine and legislative debate, but cannot establish human conception merely because patent protection is commercially desirable. Thaler, 43 F.4th at 1211–13.
IX Conclusion
AI-generated claim language does not establish an absence of human conception. A model may articulate an existing invention, help complete an operative solution, or present an unforeseen design that a human subsequently comprehends. These situations require different factual analyses. The natural-person rule excludes the model from inventor status without resolving the human’s entitlement.
The recognition account defended here treats sufficiently definite technical comprehension as a possible basis for original human conception. That application remains an argument requiring factual and doctrinal support. Project control, commercial appreciation, and approval are insufficient substitutes. The inquiry must identify the operative idea, each human’s contribution, and the disclosure available at filing.
