AI Inventorship After Thaler: MPEP §1.03(a) 2025—Name or Exclude?

This guide explains how the USPTO’s revised 2025 inventorship guidance applies after Thaler and why AI cannot replace a natural person as the inventor. It provides a claim-by-claim human-conception audit for identifying qualifying inventors and supporting evidence before filing.

How the 2025 rule actually works

Thaler v. Vidal settled the threshold identity question: an artificial intelligence system cannot be named as an inventor on a patent application. The holding does not make AI-assisted inventions categorically unpatentable. It instead leaves a familiar question for human applicants: whether the claimed invention was conceived by a natural person under the Patent Office’s governing rules.

The USPTO has issued two rounds of guidance addressing that question. In 2024, the Office stated that an invention could receive patent protection so long as a human made a significant contribution to the conception of each claim, according to BitLaw’s account of the guidance. In 2025, the USPTO withdrew that framework and replaced it with an approach that returns inventorship to traditional conception principles, principally the rules reflected in MPEP § 1.03(a). BitLaw notes that the revised guidance does not fully explain how those principles apply when an AI system originated a claimed element.

Under the revised framework, AI functions as a tool rather than an inventor—more analogous to laboratory equipment than to a member of the research team. Accelerate IP summarizes the practical rule this way: patents may issue when a natural person demonstrates meaningful human conception of the claimed invention. The human contribution need not be disconnected from AI; rather, the record must show that the person supplied the legally operative conception reflected in the claims. The 2025 approach therefore asks not whether AI was used, but whether the application satisfies the ordinary requirement that claimed subject matter be conceived by an inventor.

For each claim, the filing team should compare the claim’s limitations against the documented human inventive work. The key check is whether a named natural person can explain how the claim’s conception arose and identify the supporting notebooks, design files, model prompts, experimental records, or communications that evidence that contribution. AI output can help a human formulate, test, or refine an invention, but it should not be treated as a substitute for attributing conception to the human who is listed as inventor.

That claim-level review should be performed before filing. If the application identifies a human inventor, the file should connect that person to the conception of every claim; if the evidence instead shows only that the person directed an AI system without contributing to conception, the inventorship statement may need correction. This is the concrete consequence of the 2025 guidance: AI use does not itself disqualify the application, but the application must remain anchored to human conception under MPEP § 1.03(a).

The evidence: what the sources converge

Four independent sources converge on the same operative rule: at least one natural person must make a significant contribution to the conception of every claim. BitLaw, AIPPI, Lexology, and Accelerate IP reach that conclusion through different accounts of the USPTO’s revised inventorship guidance, but none permits a claim to rest solely on AI-originated elements. AI may assist the inventive process; the human contribution must nevertheless extend to the conception of the claimed subject matter itself, not merely to the overall project, the commercial objective, or an implementation detail.

BitLaw’s formulation supplies an important caution. The guidance “states that inventorship must be evaluated under traditional conception principles,” but it does not explain the application of those principles when an AI system originated particular claimed elements. That silence does not erase the cross-source rule; it signals where claim drafting and inventorship evidence can become difficult. A submission cannot cure a missing human conception contribution merely by adding an inventor’s name to an application.

AIPPI’s account emphasizes the practical consequence: the update “restores predictability for applicants” by anchoring analysis in a human inventor’s significant contribution to each claim. Lexology likewise describes the revised guidance as a streamlined return to a traditional framework. Accelerate IP expresses the same standard as a demonstrable showing of meaningful human conception. These are not materially different formulations of the bottom-line requirement, although their terminology helps translate it into a claim-by-claim review.

Before filing, the practical audit is therefore straightforward but demanding: for each claim, identify the named natural person who made a significant contribution to its conception and assemble documentary evidence of that contribution. Useful evidence can include dated laboratory notes, design records, technical communications, source-code history, and revision records showing that the human conceived or materially shaped the claimed limitation. The evidence should connect the individual to the claim—not merely to the invention’s general field or business purpose—and should show how the contribution affected the claimed conception rather than only later implementation. This audit is the evidence-based line between an AI-assisted invention and a claim whose conception rests solely on AI-originated elements.

Name, exclude, or narrow: options compared

Option A is to name the artificial intelligence system as an inventor. That posture is foreclosed in the United States. As BitLaw summarizes Thaler v. Vidal, an AI system cannot be named as an inventor, and the USPTO’s revised guidance for AI-assisted inventions retains that threshold identity requirement. The application would therefore be fatally defective as filed. Adding or naming a human later would not cure the defect, because the necessary judgment is whether a natural person made the required inventive contribution—not whether the filing can be repaired after the fact.

Option B is to exclude the AI from the inventor declaration and name only the participating humans. BitLaw describes two successive rounds of USPTO guidance: the Office’s earlier significant-contribution framework was withdrawn and replaced in 2025 with guidance that returns inventorship to traditional conception principles under MPEP §1.03(a). Naming only humans is viable only when every claim is supported by a documented, significant human contribution to its conception. If a claim is wholly or partly attributable to AI-generated material and no natural person made the required contribution, excluding the AI as an inventor does not solve the problem. The resulting patent would remain vulnerable to an improper-inventorship challenge and associated invalidity consequences.

Option C is to narrow the claims so that each one is limited to elements conceived through a qualifying human contribution. This is the winning filing posture because it aligns the scope of the claims with the conception record rather than asking the USPTO to treat an AI system as an inventor. The practical check is claim-by-claim: identify the named human who significantly contributed to the conception of each limitation, connect that person to the supporting records, and remove or revise any limitation for which the contribution cannot be shown.

The best practice is therefore not an unconditional choice between naming and excluding AI. For human-conceived subject matter, name the qualifying human inventors and explain their contributions. For AI-originated subject matter, narrow the claims or leave that subject matter out. Before filing, the claim-by-claim human-conception audit should answer one question for every claim: who is the named natural person, what did that person contribute to the claim’s conception, and what documentary evidence supports that contribution? Claims that cannot pass that test should not be presented as human-invented merely because a human used or supervised the AI system.

Costs and numbers that matter

The principal cost categories are: domestic inventorship correction, foreign inventorship correction, and the prosecution cost of narrowing or removing claims that lack supportable human conception. These are distinct expenses, and a filing strategy that treats them as a single “inventorship” issue can understate both the monetary exposure and the risk of losing claim scope. Before filing, the useful check is whether the file can identify the natural person who made a significant contribution to the conception of each claim and preserve evidence connecting that person’s contribution to the claim language.

Domestic correction is governed by 35 U.S.C. § 256. An incorrect inventorship declaration is correctable when the error occurred without deceptive intent, but the qualification is fact-dependent. That makes a correction process vulnerable to dispute: the applicant may need to explain the chronology of conception, distinguish each inventor’s contribution, and address whether the original designation was the result of an innocent recordkeeping error rather than an attempt to mislead the USPTO. The practical cost therefore includes more than preparing a new declaration; it can include reconstructing the invention record, collecting declarations, briefing the legal standard, and responding to scrutiny. The client-facing check is simple: before seeking correction, determine whether the evidence supports the no-deceptive-intent condition without relying on a conclusory assertion.

Foreign correction presents a different cost and a different timing problem. AIPPI notes that some jurisdictions permit later amendment to correct foreign inventorship where possible. Consequently, a defective inventorship position in a U.S.-first filing may have a correction route abroad even when the corresponding U.S. error is not correctable under § 256. That asymmetry should be investigated jurisdiction by jurisdiction rather than inferred from U.S. practice. For each relevant foreign office, confirm whether a later amendment is available, what showing it requires, and whether the corrected inventorship must be reflected in the application’s substantive record. The key comparison is between a domestic path centered on § 256 and any foreign process whose availability, timing, and evidentiary demands must be verified locally.

Narrowing creates a third category of cost: prosecution expense coupled with the loss of commercial claim scope. A claim that cannot be traced to significant human conception may have to be dropped or narrowed to the subject matter that the documented human contribution supports. Dropping a claim can leave the application with fewer requested protections; narrowing can require different claim language, support, priority treatment, and argument, while still not curing an inventorship error elsewhere in the application. The check is claim-by-claim: compare the final claim with the conception record, identify the named human contributor, and ask whether the evidence permits that contributor to be associated with every operative limitation. If it does not, price both the prosecution work required to revise the claim and the business cost of the resulting narrower protection. This is why a pre-filing human-conception audit is more useful than a last-minute inventorship label: it exposes claim-scope cost before filing decisions become difficult to unwind.

What the evidence does NOT establish

The 2025 guidance leaves the hardest part of AI-assisted inventorship unresolved: how traditional conception principles apply when an AI system originated some of the elements claimed in a patent application. BitLaw identifies this mixed-origin problem as a central uncertainty in the revised framework. A human may direct the overall inventive process, yet still need to establish a significant contribution to the conception of each specific claim. The presence of human involvement somewhere in the project does not, without more, show that the human conceived every claimed limitation.

That creates a claim-by-claim audit problem. Before filing, identify the natural person who made a significant contribution to the conception of each claim, and preserve evidence connecting that person’s work to the claimed limitations. The useful question is not whether a human merely supervised an AI system or approved its output. The question is whether the human contributed to the conception of the claim as claimed, and whether the file can support that conclusion with dated drafts, instructions, design records, experimental notes, or communications with the relevant system and coworkers.

None of the identified sources supplies a threshold defining how large a human contribution must be. BitLaw, AIPPI, Lexology, and Accelerate IP describe the governing “significant contribution” inquiry, but the available sources does not establish whether prompting an AI system, selecting among generated outputs, or curating training data can itself qualify. Nor does it establish a rule that one of those activities is sufficient. Each activity may matter in context, depending on what the human actually supplied and whether the contribution concerned conception rather than routine implementation or reduction to practice.

The guidance likewise provides no numerical test, percentage, weighting formula, or other measurement for comparing human and AI contributions. Any such percentage would be invented. The defensible approach is a qualitative record: document what the human proposed or changed, what the AI produced, how the human evaluated the output, and why the final claim limitation reflects a human inventive contribution. For each claim, the file should connect a named human to a significant conception contribution through contemporaneous evidence.

The resulting edge cases are precisely where caution is warranted. Examples include a claim combining an AI-generated component with human-designed architecture, a human refining an AI proposal without directing its initial generation, and a claim selected from several machine-generated alternatives. The guidance does not resolve how to evaluate these situations categorically. Accordingly, the claim-by-claim human-conception audit should identify not only the proposed inventor but also the claimed limitation attributable to that person, the supporting documents, and any unresolved attribution issue that may require narrowing the claim or obtaining further evidence before filing.

The claim-by-claim audit

Before filing, I run a claim-by-claim human-conception audit. The threshold is not whether AI appears in the invention record; it is whether a named natural person made a significant contribution to the conception of each claim. The 2024 USPTO framework, which expressly used that significant-contribution formulation, was withdrawn and replaced in 2025 with revised guidance that returns inventorship to traditional conception principles. Thaler v. Vidal established that an AI system cannot be named an inventor, but it did not categorically bar patent protection for inventions made with AI assistance. The 2025 guidance leaves the practical question: can the human contribution to each claim be identified and documented?

For this audit, work through each claim in the application and record, for every claim, the named natural person who made a significant contribution to its conception and the documentary evidence supporting that contribution. The worksheet below shows how that review plays out for three illustrative claim types.

Apply the same three questions to every claim: who is the named natural person, what did that person contribute to the claim’s conception, and what documentary evidence supports that contribution? A claim whose conception rests solely on AI-originated elements should be dropped or narrowed to subject matter supported by a documented human contribution. A claim supported by a documented human conception contribution may be kept, with the supporting rationale preserved in the invention record. Where the human contribution to the claimed subject matter cannot be documented, narrow or drop the claim rather than presenting it as human-invented.

The audit therefore produces a concrete result for any claim whose conception rests solely on AI-originated elements: drop or narrow it to subject matter supported by a documented human contribution. AI-generated output can be part of the work product, but it cannot substitute for the required human conception record. Where a named human made a significant contribution to a claim’s conception, keeping that claim requires more than adding an inventor name; the invention record should preserve the rationale, experimental basis, and contemporaneous evidence supporting the contribution.

For any claim involving AI optimization, the audit should separate the AI’s activity from the human’s claimed contribution. A human who merely accepted the system’s recommendation, or merely approved a result without selecting or shaping the claimed subject matter, may not supply the required conception. The relevant question is what the human contributed to the claimed invention, and the file should show it claim by claim.

The filing posture follows from the audit rather than from a slogan. Naming AI is unavailable after Thaler because AI cannot be an inventor. Excluding AI entirely may unnecessarily discard patentable subject matter, since the 2025 guidance permits AI to function like a tool. Narrowing claims is the winning posture where the application otherwise lacks a documented human contribution: it preserves claims that satisfy the human-conception threshold and removes claims that do not. BitLaw, AIPPI, Lexology, and Accelerate IP converge on the practical requirement of at least one human contribution for every claim, even when AI participated in the inventive process. The final worksheet should therefore identify, for every claim, the named human and the documentary evidence of that contribution before filing.

Decision rules: five if/thens

The practical framework changed twice before the current rule emerged. In 2024, the USPTO stated that an invention could qualify for patent protection when a human made a significant contribution to the conception of each claim; in 2025, the Office withdrew that framework and replaced it with guidance grounded in traditional conception principles. BitLaw notes that the revised guidance leaves some uncertainty where AI originated claimed elements. The threshold was settled in Thaler v. Vidal: an AI system cannot be a named inventor. Under MPEP §1.03(a) lineage, the filing decision therefore turns on whether a natural person can be identified and connected to the conception of each claim.

First, if a claim’s conception is traceable to a named human and supported by documentary evidence, then keep the claim and name that human as an inventor. The audit should connect the claim to the human’s specific contribution—for example, the design, arrangement, or technical relationship recited in the claim—and preserve the notebooks, drawings, communications, dated versions, or other contemporaneous materials showing that contribution. Merely naming a person because they supervised the project is not enough.

Second, if a claim’s conception originated with the AI and no human made a significant contribution to that claim, then drop or narrow the claim, and do not name the AI. A useful audit question is whether any required claim limitation reflects a human’s conception of the claimed solution. If the record cannot connect a human to that limitation, merely adding AI-assisted wording will not cure the deficiency.

Third, if the human contribution consists only of prompting, treat it as insufficient until the record establishes a significant contribution, and narrow the claim accordingly. Asking an AI for alternatives may produce useful work product, but the documentation must show that the human supplied, evaluated, or further developed a claimed inventive concept. A prompt alone is not a substitute for claim-specific evidence.

Fourth, if the asserted human contribution is only data selection, treat it as insufficient until documented otherwise, and narrow the claim. Selecting source material or a training corpus may explain what the AI processed, but it does not necessarily establish conception of the claimed invention. The file should connect the human to the claimed structure, relationship, or mechanism rather than to the data used by the system.

Fifth, if the asserted human contribution is only output curation, treat it as insufficient until documented otherwise, and narrow the claim. Choosing among AI-generated outputs is not automatically a significant contribution to conception. Before retaining a claim on that basis, document the human judgment that shaped or completed the claimed concept—for example, resolving a technical problem, selecting a particular arrangement, and explaining why that arrangement satisfies the claim. Accelerate IP’s formulation is a practical check: AI may function as a tool, but a natural person must be able to demonstrate meaningful human conception. Before filing, complete this analysis claim by claim, record the named contributor and the supporting evidence, and revise claims that fail the audit.

What to do next

Frequently Asked Questions

Can an AI system be named as the inventor on a USPTO patent application?

No; MPEP §1.03(a) prohibits naming AI as an inventor.

Does Thaler v. Vidal make every AI-assisted invention unpatentable?

No; the holding addresses whether AI can be an inventor and does not categorically bar patents for AI-assisted inventions.

How many natural-person inventors must be identified for each claim?

At least one natural person must be named as inventor for every claim.

What contribution must a human inventor make to a claimed invention?

The human inventor must make a significant contribution to the conception of each claim.

What should be documented before filing a patent application involving AI-generated elements?

Document the natural person who significantly contributed to the conception of each claim and the evidence supporting that contribution.

What should be checked when an invention contains AI-generated elements and may be filed abroad?

Check the applicable foreign inventorship rules, including whether foreign inventorship can be corrected through later amendment.

Quick answers

Also worth reading: M&A Recordal: USPTO 2 Weeks vs EPO 6 Months—No Mere Formality: M&A Recordal: USPTO 2 Weeks · Design Patent Priority: The Six-Month Clock Under Paris 4(C): Design Patent Priority: The Six-Month · Weight loss drug patents: 2024-2025 longevity indications hit 37 — watch or challenge?: Weight loss drug patents: 2024-2025