Who Owns the AI-Assisted Invention? A Practical Guide for Businesses Using AI in R&D

Your engineering team just used an AI model to help design a novel drug-delivery mechanism. Another engineer prompted a generative AI system to explore an alternative circuit topology, and the AI’s output pointed directly to the solution your team ended up patenting. A contractor working remotely used the company’s enterprise AI license to develop a new algorithm that now sits at the core of your product. In each scenario, a human being hit “submit,” evaluated the output, and built something real from it. But a fundamental question hangs over all of it: who actually invented the thing, and who owns the patent?

Those are not the same question, and the gap between them is where businesses get into serious trouble. This post walks through what the law currently requires, what the USPTO now says about AI-assisted inventions, and what your company needs to do right now to protect the IP your teams are generating.

The Foundational Rule: AI Cannot Be an Inventor

Start with the bedrock principle, because everything else follows from it. In Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), the Federal Circuit held that the Patent Act’s use of the word “individual” requires that an inventor be a human being. The court looked at the ordinary meaning of “individual” as a noun — a human being, a person — and concluded that AI systems are categorically excluded from inventorship. The Supreme Court declined to hear the case, leaving that holding intact as settled law.

This means no matter how autonomously an AI system generates a solution, no matter how minimal the human’s role appears in the moment, the AI cannot be named as inventor on a patent application. Only the humans who contributed to the conception of the claimed invention can be inventors. If no human made the requisite contribution, there is no patentable invention — the subject matter falls into the public domain.

For businesses, this ruling is both a constraint and an opportunity. It is a constraint because it requires that human engineers and scientists be meaningfully involved in the conception of claimed inventions, not just in prompting an AI and accepting its output. It is an opportunity because it means your company can still obtain valid patents on AI-assisted work, provided the humans involved qualify as inventors under the applicable legal standard.

The USPTO’s Current Guidance on AI-Assisted Inventions

The USPTO has issued guidance on this topic twice in recent years, and understanding the current version matters for anyone filing patent applications today.

The February 2024 Guidance and Its Fate

On February 13, 2024, the USPTO published “Inventorship Guidance for AI-Assisted Inventions” in the Federal Register, 89 FR 10043. That guidance centered its analysis on the three-factor test from Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998), requiring each human contributor to demonstrate that they made a significant contribution to the conception of the claimed invention. The 2024 guidance asked examiners and applicants to evaluate AI use against the Pannu factors even where only a single human was involved. It drew criticism from portions of the patent community for importing a joint-inventorship framework into situations where there was no joint inventorship question to analyze, since the AI cannot be a co-inventor.

The November 2025 Revised Guidance: What It Actually Says

On November 28, 2025, the USPTO rescinded the February 2024 guidance entirely and replaced it with revised guidance published at 90 FR 54636, Docket No. PTO-P-2025-0014. The revised guidance is now the operative USPTO policy, and businesses filing patent applications need to understand it.

The core principle of the 2025 guidance is straightforward: the same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used. There is no separate or modified standard for AI-assisted inventions. This represents a deliberate simplification. The USPTO made clear that AI systems — including generative AI and other computational models — are instruments used by human inventors, analogous to laboratory equipment, computer software, or research databases. The Federal Circuit said as much in Shatterproof Glass Corp. v. Libby-Owens Ford Co., 758 F.2d 613 (Fed. Cir. 1985): inventors “may use the services, ideas, and aid of others” without those sources becoming co-inventors.

The Conception Standard in Practice

The 2025 guidance centers the entire inventorship inquiry on conception, which the Federal Circuit in Burroughs Wellcome Co. v. Barr Labs., Inc., 40 F.3d 1223 (Fed. Cir. 1994), described as “the touchstone of inventorship.” Conception is “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.” Conception is complete when the inventor has “a specific, settled idea, a particular solution to the problem at hand, not just a general goal or research plan.”

Put practically: a human engineer who reviews an AI output, evaluates its technical feasibility, recognizes that it solves the problem, selects among competing AI-generated options, and understands why the solution works is likely contributing to conception. A human who simply prompts an AI with a broad problem statement and adopts whatever the AI produces without meaningful technical evaluation is in a weaker position — not necessarily disqualified, but exposed to the argument that the human never possessed “a definite and permanent idea of the complete and operative invention” as required by the Federal Circuit’s formulation.

When Multiple Humans Are Involved

When multiple natural persons collaborate on an AI-assisted invention, the Pannu factors apply between the human contributors in the same way they always have for joint inventorship. Each person must: (1) contribute in some significant manner to the conception or reduction to practice of the invention; (2) make a contribution that is not insignificant in quality when measured against the full invention; and (3) do more than merely explain well-known concepts or the current state of the art to the real inventors. The AI’s role does not alter this analysis among humans — it just means you evaluate the humans against each other, not against the machine.

Why Getting Inventorship Wrong Is Genuinely Dangerous

Inventorship is not a technicality. It is a legal condition that affects the validity, enforceability, and ownership of a patent. Getting it wrong — whether by naming someone who is not a true inventor, or by omitting someone who is — creates serious vulnerabilities.

Patent Invalidity

Under 35 U.S.C. § 115, patent applications must correctly identify the inventor or joint inventors. An issued patent with an incorrect inventorship is subject to challenge. Where a patent omits a true inventor and the error cannot be corrected, the patent is invalid. Courts have rendered patents unenforceable and dismissed infringement suits on this basis.

The Correction Mechanism — and Its Limits

Congress created a safety valve in 35 U.S.C. § 256, which allows the Director of the USPTO to issue a certificate correcting inventorship when an inventor was added or omitted through error. Under § 256(a), the USPTO can issue a correction upon application; under § 256(b), a district court may order such a correction during litigation. Section 256 is a “savings provision” designed to prevent invalidation of patents due to good-faith errors.

But § 256 has limits. Courts have held that corrections can be barred where there was deceptive intent — that is, where the incorrect inventorship was not an accident. More significantly, a recent line of cases has established what is sometimes called “uncorrectability” as a patent invalidity defense: where an omitted co-inventor exists but cannot satisfy the procedural requirements of § 256, the patent may be rendered invalid in litigation. If an omitted inventor cannot be located, refuses to cooperate, or the error appears intentional rather than inadvertent, correction may be unavailable — and the patent along with it.

Inequitable Conduct

There is also the threat of inequitable conduct, which renders a patent unenforceable (not just invalid). To establish inequitable conduct, a challenger must show that a patent applicant made a misrepresentation or omission of material information with specific intent to deceive the USPTO. In the inventorship context, deliberately listing someone as an inventor who did not conceive any portion of the claimed invention, or deliberately omitting someone who did, can support an inequitable conduct claim. An unenforceable patent cannot be asserted against infringers and can be used as a basis for fee-shifting in litigation.

In the AI-assisted context, the risk is concrete: if your team includes a name on the inventor list to satisfy an executive or contractual partner, without that person having made a genuine contribution to conception, you have created a fraud-on-the-USPTO problem that may surface years later during litigation.

Employee Invention Assignments and AI Assistance

For most technology companies, the practical answer to “who owns the patent?” starts with the employment relationship. Standard employment agreements include an invention assignment clause — often called an Employee Invention Assignment and Confidentiality Agreement (EIACA) — requiring employees to assign to the company all inventions made during employment, using company resources, or related to the company’s business.

The Assignment Mechanism Still Works

AI assistance does not break the basic assignment chain. If an employee uses the company’s AI tools to help conceive an invention, and that employee signs a proper invention assignment agreement, the company still ends up owning the patent through assignment from the named human inventor. The fact that an AI tool was involved in generating ideas or exploring solution spaces does not alter who the human inventor is, and it does not alter the employee’s obligation to assign.

However, invention assignment agreements written before the generative AI era may have gaps. Older agreements were often drafted with a human-only conception framework in mind. They may not explicitly address inventions developed using AI tools, or they may condition the assignment obligation on inventions “conceived by the employee,” which a clever adversary might argue is ambiguous when an AI contributed heavily to the idea.

What Employment Agreements Should Address Today

Companies should review their EIACA language to confirm that it covers inventions developed with the assistance of AI tools. The agreement should make clear that: (a) the use of any software, AI system, or computational tool in the development of an invention does not change the employee’s assignment obligation; (b) the employee represents that they made a genuine human contribution to the conception of any inventions they disclose; and (c) the obligation to disclose extends to AI-assisted work product generated using company systems or during work hours.

Several states — including California, Washington, Delaware, Illinois, and Minnesota — limit the scope of invention assignment clauses for inventions made entirely on the employee’s own time, without using company resources, and not related to the company’s business. AI-assisted inventions developed using the company’s AI subscriptions and tools will generally fall outside these carve-outs, since company resources are being used.

Contractor and Consultant Ownership: The Gap That Bites Companies

The contractor situation is fundamentally different from the employee situation, and the difference is critical. Unlike employees, independent contractors own their inventions by default. There is no “work made for hire” doctrine for patents — that doctrine applies to copyright, not inventorship. In the patent world, the only way a company acquires ownership of a contractor’s invention is through a written assignment.

Why “Work for Hire” Does Not Solve the Patent Problem

Many business owners assume that if they paid for something, they own it — especially if there is a “work made for hire” clause in the services agreement. That is broadly correct for copyright (subject to the nine statutory categories under 17 U.S.C. § 101), but it does not apply to patents. Patent ownership follows inventorship. The inventor owns the patent unless the inventor has agreed in writing to assign it. A work-for-hire clause in a services agreement will not transfer patent rights.

AI Tools Make the Contractor Problem Worse

When a contractor uses your company’s AI license to develop an invention, the ownership question becomes more complicated, not less. The contractor’s human contribution to conception may qualify them as the inventor. Their use of your AI tool does not automatically transfer the resulting invention to your company. Without a written assignment, the contractor may own a patent on work your company paid for and that was generated using your infrastructure.

The fix is straightforward but must be done in advance: every services agreement involving any technical or inventive work should include an explicit IP assignment clause transferring all inventions — including AI-assisted inventions — to the company, along with a clause requiring the contractor to execute any documents necessary to perfect that assignment (patent applications, USPTO assignments, continuations). Requiring assignment “of all inventions, discoveries, and improvements made or conceived, alone or with others, in connection with the performance of services” covers the relevant ground.

Contractor Agreements Should Also Address Disclosure

Contractors should be required to disclose inventions promptly and to cooperate in the patent application process, including execution of inventor declarations. If a contractor later becomes unavailable or uncooperative, 35 U.S.C. § 116 allows the other inventors or the applicant company to proceed in some circumstances, but you want the contractual obligation in place before that situation arises.

AI Vendor Terms of Service: Does Using ChatGPT or Copilot Affect Who Owns the Patent?

This question generates more concern than it deserves, but it deserves a clear answer. When your engineers use a third-party AI tool — OpenAI’s API, GitHub Copilot, Google’s Gemini, or any commercial AI platform — to assist in R&D, do the vendor’s terms of service create any ownership rights in the resulting invention?

What the Major Vendors Actually Say

OpenAI’s terms of service state that users retain ownership of their inputs and that OpenAI assigns to the user all of its right, title, and interest in the outputs. OpenAI does not claim a license over inputs or outputs, and has stated publicly that it will not claim copyright over content generated through the API. GitHub’s terms take the same posture — GitHub states it does not claim ownership rights in user content.

These provisions address copyright, not patents. But the practical effect is the same: the major commercial AI vendors are not claiming any ownership interest in inventions that their tools help create. Their terms of service do not purport to acquire patent rights based on a user’s use of the service.

The More Realistic Risk: Training Data and Prior Art

The vendor terms raise a different concern that is more practically relevant: confidentiality. Most AI services, unless you are on an enterprise tier with explicit data-isolation provisions, may use your inputs to improve their models. If you are entering proprietary technical information — formulas, circuit designs, drug candidates, novel process parameters — into an AI prompt without enterprise confidentiality protections, you may be creating disclosure risks that affect your ability to obtain patents. Public disclosure of an invention before filing triggers the one-year bar under 35 U.S.C. § 102(b)(1). Even non-public disclosures to third parties can complicate foreign patent filings where absolute novelty is required.

The practical rule: use enterprise AI agreements with explicit data confidentiality provisions for any R&D involving potentially patentable subject matter. Read the terms before using consumer-grade AI tools for technical development work.

Open Source and Copilot-Specific Concerns

GitHub Copilot is trained on publicly available code, including open source code. The terms put the compliance responsibility on users: if Copilot suggests code that matches code in its training set, the user is responsible for ensuring compliance with applicable licenses. This is primarily a copyright and open source licensing issue, not a patent inventorship issue, but it is worth noting that accepting a Copilot suggestion verbatim is not the same as a human engineer conceiving the underlying technical solution independently.

Documentation Protocol for AI-Assisted R&D

The most important thing your R&D teams can do today — before you file a single patent application — is implement a documentation protocol that creates a contemporaneous record of human conception. Here is what that looks like in practice.

Invention Disclosure Forms

Every potential patent should begin with an invention disclosure form (IDF) submitted to patent counsel or the company’s IP committee. In the AI-assisted context, the IDF should capture not just what the invention is, but how it was conceived. Specifically:

• What problem was the inventor trying to solve?
• What AI tools were used, and what prompts or inputs were provided?
• What did the AI generate in response?
• What did the human inventor evaluate, select, modify, or reject from the AI’s output?
• What technical judgment did the inventor apply in recognizing that the AI output solved the problem and was workable?
• What aspects of the claimed invention did the human develop independently of the AI output?

This narrative is the evidentiary foundation for inventorship. It documents the human’s mental engagement with the invention — the formation of “a definite and permanent idea of the complete and operative invention” that the Federal Circuit requires.

AI Interaction Logs

Many enterprise AI platforms generate logs of prompts and responses. These logs should be preserved as part of the R&D record for any invention that proceeds to patent prosecution. They serve two purposes: they help confirm the human’s analytical contribution (by showing what the human did with the AI’s output), and they establish the timeline of conception if priority disputes arise.

Electronic Lab Notebooks

For research-intensive organizations, electronic lab notebooks (ELNs) with timestamped, authenticated entries are the contemporary equivalent of the traditional signed and witnessed paper lab notebook. Entries should document the inventor’s thought process, the AI tools used, the outputs reviewed, and the engineering decisions made. The record should be specific enough that, if the patent is challenged years later, your inventor can sit in a deposition and walk through exactly how the conception occurred.

Inventorship Review Before Filing

Every patent application involving AI-assisted development should go through an inventorship review before filing. This means patent counsel should interview each potential inventor to establish what that person’s specific contribution to conception was, as against the individual claims being pursued. In the AI-assisted context, the review should also identify whether anyone was primarily a conduit for AI output (and thus potentially not a true inventor) and whether anyone who made genuine contributions has been inadvertently omitted.

Document the results of the inventorship review in writing. If counsel concludes that certain people should not be named as inventors, or that others should be added, that determination and its basis should be memorialized. This documentation is your defense against an inequitable conduct allegation if the inventorship is later questioned.

Joint Development Agreements and AI: Complications for Multi-Party Projects

Technology companies increasingly co-develop products with partners, research institutions, and other companies. Joint development agreements (JDAs) govern who owns the resulting IP, but most JDAs in active use were not drafted with AI-assisted development in mind.

The Default Rule Is Inconvenient

Under 35 U.S.C. § 262, absent an agreement to the contrary, each co-owner of a patent may make, use, sell, and license the invention without accounting to the other co-owners. This rule exists because requiring all co-owners to consent to every license would create holdout problems. But in a joint development context, this rule means that if your partner’s engineer contributes to conception and your company cannot establish clear ownership through the JDA, your partner could independently license the jointly-owned technology to your competitors.

How AI Use Complicates Inventorship in Joint Development

In a typical joint development project, inventorship is determined by which natural persons contributed to the conception of each claimed invention. AI tools used in the project do not automatically become “joint owners” of anything (they cannot be), but they can obscure the question of which human contributed what.

Consider: your company provides the AI tools and the technical prompts; the partner’s scientists evaluate the outputs and design the final solution. Or the reverse: the partner provides a proprietary AI model; your engineers use it to solve a problem you brought to the collaboration. In each case, the human contribution to conception is the relevant question, but the AI’s role may make it genuinely difficult to parse whose mental contribution was decisive.

Well-drafted JDAs should address these issues explicitly. Key provisions include:

Ownership allocation of AI-assisted inventions. The JDA should specify how inventions developed using AI tools — and particularly AI tools provided by one party but used by both — will be owned. Options include automatic assignment to the party whose employees are the named inventors, joint ownership with specific licensing rights, or a technology committee that reviews and allocates inventions as they arise.

Background IP and AI tools. If one party is contributing proprietary AI systems as background IP, the JDA should address whether inventions generated with the assistance of those AI tools are considered jointly developed IP or the exclusive IP of the providing party. A poorly drafted provision can inadvertently give a partner ownership rights in output generated by your proprietary AI.

Inventorship review process. The JDA should establish a procedure for jointly reviewing inventorship determinations before patent applications are filed. In practice, this means requiring each party to disclose potential inventions to a joint IP committee, with an agreed-upon timeline for determining inventorship and filing decisions.

Documentation obligations. Each party should be required to maintain records documenting their personnel’s contributions to the conception of inventions, including records of AI tool use. These records may become critical if inventorship is disputed — and in joint development relationships, inventorship disputes are not uncommon.

When to Bring in Patent Counsel

If your company is engaged in any of the following, you need patent counsel involved before applications are filed, not after:

• R&D teams regularly using generative AI tools to explore technical solutions
• Contractors or consultants doing inventive technical work with company AI systems
• Active joint development relationships without IP provisions drafted in the last two years
• Employment agreements older than three years that have not been updated to address AI-assisted work
• Product development pipelines where AI tools are generating outputs that inform patentable claims

The cost of an inventorship problem discovered during patent litigation — invalidity challenges, inequitable conduct claims, co-inventor disputes — vastly exceeds the cost of getting the documentation and agreements right at the outset.

Conclusion

The legal framework for AI-assisted invention ownership is settled in its foundations: only human beings can be inventors, the same conception standard applies regardless of AI involvement, and ownership flows from inventorship through assignment agreements. What is unsettled is how well most companies have adapted their internal processes to this reality.

The USPTO’s November 2025 guidance makes the standard clearer and more workable: treat AI as a tool, apply the traditional conception analysis to the humans involved, and document the human contribution to each claimed invention. The businesses that will protect their R&D investment are those that treat inventorship review not as a formality to be handled at the last minute before filing, but as an integral part of the development process — one that begins when the AI tool is first engaged and ends only when the patent issues.

Update your employment agreements. Audit your contractor agreements. Negotiate AI-specific IP provisions into every joint development relationship. Train your R&D teams on what it means for a human to genuinely conceive an invention when AI tools are in the room. And document, document, document.

The patent system was designed to reward human ingenuity. The question is whether your documentation proves that the ingenuity was yours.

This post is for general informational purposes and does not constitute legal advice. Patent law is highly fact-specific. Consult a qualified patent attorney regarding your company’s particular circumstances.