What an AI patent inventorship audit actually determines
An AI patent inventorship audit is a documented review of who contributed to the conception of an invention, including contributions made through an AI system. Its purpose is not to determine whether a patent application should be filed or whether an invention is valuable. Instead, it tests whether the proposed inventors satisfy the applicable inventorship rules, whether omitted or improperly named inventors could create a validity problem, and whether the company has evidence supporting its ownership and authorization decisions. In the United States, inventorship turns on the contribution to conception, not on who directed a project, funded the research, supplied data, or merely described a desired result. That distinction becomes difficult when a human posed the problem, selected the model, supplied prompts, evaluated outputs, and selected which output became the invention. As of October 2, 2026, companies should not assume that prompting an AI makes the prompt author the sole inventor. The audit should reconstruct each claimed feature and trace it to the human or humans who conceived that feature. It should also distinguish conception from routine implementation: naming a parameter, arranging conventional components, or telling a skilled person what to build generally does not make a person an inventor of the entire result. An audit does not cure naming errors automatically. If incorrect inventorship appears, counsel may need to correct a naming record or decide how to proceed under the governing application status and procedural rules. The safest audit is performed before filing, but it can still be useful after filing or after receiving an office action.
Also worth reading: AI Patent Inventorship Records: Who Must Be Named on AI-Assisted Patents in 2026? · What Proves AI Inventorship in a U.S. Patent Application in 2026? · How Do AI Patent Filing Controls Affect Inventorship, Disclosure, and Filing Strategy?
The governing rules for human and AI-assisted inventions
The U.S. patent system requires the inventor or inventors to be the natural person or persons who conceived the claimed invention. AI systems are not inventors under current U.S. practice, so a company must identify the qualifying natural-person contributions. USPTO guidance addresses AI-assisted inventions by asking whether a human contributed significantly to the claimed subject matter, rather than merely providing a prompt that specifies the claimed invention with enough detail that little or no inventive work remains. A human who conceives a solution using conventional tools may qualify even where the tools materially assist drafting or analysis. Merely supplying data, presenting an idea already conceived by someone else, or requesting a particular output ordinarily does not establish inventorship by itself. The analysis is claim-specific. Different people can be inventors of different claims, and a person need not contribute to every limitation if that person conceived at least one principal feature of the claimed combination and made the contribution to conception. Conversely, a person whose work is confined to implementation is not an inventor merely because the implementation was technically demanding. No universal percentage test states how much effort makes someone an inventor, and no universally applicable threshold converts a 20%, 50%, or 80% contribution into legal inventorship. Inventorship is legal, not arithmetic. The audit should therefore document the actual sequence of technical decisions and avoid labels such as “primary inventor” unless the underlying contributions are legally supportable. International rules can differ, making a jurisdiction-by-jurisdiction review necessary when filings are planned worldwide.
A claim-by-claim audit method
The audit begins by defining the version of the invention being reviewed, including the date, repository, model name, model version, prompt history, source data, and attached documents. Counsel should then prepare a claim chart that maps every claim element to evidence of conception. For each element, the team records who first proposed the operative concept, what alternatives were considered, why the selected approach was chosen, and whether the AI merely transformed an already complete instruction. Interviews may be necessary because commit histories and prompt logs often reveal what was produced but not who conceived the protected idea. Contributors should be questioned separately to reduce group influence, and each answer should be tied to notebooks, design documents, tickets, source code, or dated messages. The evidence should be contemporaneous where possible. Reconstructed declarations written years later are weaker than records created at the time, particularly when the goal is to exclude an employee or contractor. A useful audit also identifies inventorship changes over time. A first draft may reflect one conception, while a later revision introduces a new feature conceived by a different person. If new matter is added, counsel must assess whether the revision remains within the original disclosure. The final report should separately state the factual findings, the legal analysis, any unresolved factual questions, and recommended correction procedures. It should not simply ask managers to certify that each contributor wrote at least 50% of the code.
Human and AI contributions compared
No table can decide inventorship by itself, but a structured comparison helps prevent marketing language from substituting for legal analysis. The relevant inquiry is the human contribution to conception of the claimed invention and the role of the AI as a tool, source, or transformation mechanism. This comparison should be tailored to the evidence rather than treated as a safe harbor or disqualification rule.
| Feature | Human-directed use of AI | Human conception followed by AI implementation | Autonomous or unexplained AI output |
|---|---|---|---|
| Core issue | Whether the human supplied the inventive concept or substantially selected and refined the claimed combination | Whether the human already conceived the claimed subject matter before using AI to draft, code, or model it | Whether a natural person can be identified as having conceived the claimed features |
| Possible inventorship | One or more qualifying humans, claim by claim | The qualifying human or humans, subject to the actual contribution | No legal AI inventorship; investigation may still identify a qualifying human |
| Useful evidence | Dated problem statements, design decisions, experiments, and revisions | Early human notebooks, specifications, sketches, and the AI request history | Model settings, logs, outputs, selection records, and the basis for adopting the result |
| Common error | Assuming a long prompt proves conception | Assuming a draft is never inventive because AI wrote it | Assigning inventorship to the developer of a general model |
| Recommended treatment | Reconstruct the human contribution element by element | Record the human conception before generation | Pause the filing decision and investigate contribution |
Evidence, records, and governance controls
A defensible AI inventorship audit depends on records that show how the claimed invention was conceived. At minimum, a company should preserve the exact model and system version, prompts, retrieval sources, system instructions, generated outputs, human edits, discarded alternatives, testing records, and decisions approving the claimed configuration. If the AI used external data or retrieved documents, the team should record what information was available and whether the output copied, combined, or improved it. The audit should also distinguish an AI tool from a human technical service, because the legal role of a supplier's personnel is different from the role of a company's employee. Contractors, consultants, university laboratories, and joint-development partners may create separate ownership and inventorship questions. Invention-assignment agreements should be reviewed alongside the inventorship analysis. An agreement can transfer rights in what a person invents, but it cannot make that person an inventor if the person did not legally invent the subject matter; likewise, inventorship is not automatically fixed by an employment title. For high-value inventions, organizations can require a short contribution declaration before filing, a second-person technical review of the contribution map, and counsel review whenever an AI system participated in a material design. They should also set a retention period long enough to cover the life of the resulting patent portfolio and any anticipated dispute, which can extend for many years after the filing date.
Mistakes that can invalidate the review
The most common mistake is treating authorship, funding, supervision, or coding volume as equivalent to conception. A senior executive who approved a budget is not necessarily an inventor, and a junior engineer who implemented every instruction is not necessarily an inventor merely because the engineer wrote most of the code. Another error is asking the AI to produce a final claim set and then naming the person who submitted the application without analyzing who conceived the technical features. A third error is treating model developers as inventors of every output generated by their system. General model development may support a separate patent application, but it does not automatically establish inventorship of a later, domain-specific result. Companies also make the mistake of running one audit across all jurisdictions without considering that some countries treat employee or academic inventorship differently and may have different rules for computer-implemented inventions and AI-assisted work. Naming a natural person for every claim may overstate inventorship, while naming only the project manager may omit a genuine inventor. A particularly serious error is relying on a retroactive declaration that merely repeats the desired conclusion. Each material statement should be tied to dated evidence and should explain what the contributor actually conceived, not how valuable the contribution was. Finally, correcting inventorship should never be treated as a clerical cleanup. The legal effect depends on the stage of the application and the applicable prosecution and equity rules, so an error should be escalated to patent counsel.
When to conduct the audit and how much it may cost
The best time to conduct the audit is before a nonprovisional or PCT filing, but the timing should reflect the value and complexity of the invention. A preliminary internal review may take several hours for a straightforward application, while a contested audit involving multiple models, hundreds of claim elements, contractor contributions, and uncertain conception history can require dozens or hundreds of hours of attorney and technical work. A focused pre-filing review by experienced patent counsel is often more economical than attempting to reconstruct evidence after a validity challenge. Companies should obtain a scoped fee estimate that states whether it includes one jurisdiction, multiple claim sets, employee interviews, model-log analysis, a written report, and a filing recommendation. The cost should be compared with the value of the patent family, the cost of correcting later, and the risk of losing rights or facing a validity dispute. There is no government filing fee for conducting an internal inventorship audit, and inexpensive software may help organize records, but it cannot perform the legal conclusion reliably. An audit is not essential for every minor invention or every use of an off-the-shelf tool. It becomes justified when AI participated in a novel technical design, more than one person may have conceived different claims, external parties were involved, or a company intends to rely on the patent in litigation, licensing, financing, or acquisition diligence. Waiting until after a notice of allowance or patent issuance can reduce procedural options, and waiting until litigation begins can make contemporaneous evidence difficult to obtain.
The recommended practical response
A company should start by assigning a responsible patent attorney and a technical lead who together identify the claimed invention and the complete generation history. They should preserve records before discussing responsibility, interview each possible contributor using the same factual questions, and separate statements about conception from statements about implementation, funding, and project management. The team should build a claim chart, identify missing evidence, and prepare a written inventorship determination for each jurisdiction in which protection is sought. If the result is uncertain, the company should not force a conclusion based on a percentage rule or on the employee's job title. It should instead investigate the underlying facts and consider narrowing the filing to claims whose conception is well supported. For ongoing innovation, the company can create a lightweight intake process requiring every AI-assisted invention to identify the model, inputs, human decisions, and external contributors. A quarterly sample review can test whether the process is capturing useful evidence, while a formal audit can be triggered for high-value or disputed matters. This approach is more demanding than naming the prompt author, but it is more consistent with current inventorship law and better suited to patent prosecution, ownership diligence, and later enforcement. The audit should end with a dated report, evidence index, identified risks, recommended naming, unresolved questions, and a plan for updating the determination when the application or claim set changes.