AI in the Patent Process: Where It Shows Up, and How to Think About It

Back to News & Insights

The term “AI” has come to mean a lot of things. In the patent process, it can show up in three different places: in coming up with the invention, in the invention itself, and in preparing and filing the application. It might show up in one of the three, in all of them, or in any combination.

Each raises a different patent question, and the United States Patent and Trademark Office (USPTO) answers each one in a separate guidance document. Mixing the three together is where most of the confusion about “patenting AI” comes from.

All three answers arrived over the past two years, and the inventorship answer changed substantially at the end of 2025. The short version is good news: the rules are simpler than they were a year ago, and they are the same rules that have always applied.

This guide takes them one at a time, based on where the AI shows up.

Our attorneys hold engineering and science degrees and focus on patent prosecution and intellectual property strategy for technology-driven companies.

Call NK Patent Law at (919) 348-2194 or contact us online to discuss your patent strategy.

Where does the AI show up?

Start by locating the AI in your own project. It can show up in several places at once: an engineer used an AI coding assistant on the firmware, the product relies on a model to do its job, you documented the invention disclosure using AI, and your attorneys used an AI tool to help draft the application. Each placement gets its own analysis.

An AI company whose engineers use AI tools and whose counsel drafts with AI hits all three at once. The rules for each are independent, which is why it is helpful to understand the difference and to keep them separate.

Our attorneys hold engineering and science degrees and focus on patent prosecution and intellectual property strategy for technology-driven companies. Call NK Patent Law at (919) 348-2194 or contact us online to discuss your patent strategy.

When AI helped your team invent: AI is a tool, not an inventor

In 2019, a computer scientist named Stephen Thaler filed two patent applications naming his AI system, DABUS, as the sole inventor. No human inventor at all. The USPTO refused to examine the applications, and Thaler appealed. The Federal Circuit held in Thaler v. Vidal that the Patent Act requires an inventor to be a natural person. An AI system cannot be named as an inventor.

Thaler is an extreme case, so it is tempting to dismiss it as absurd. Nobody at your company is trying to name a model as the inventor. The realistic concern is much more common: your firmware engineer had a coding assistant write half the driver, your RF team used generative design to explore antenna geometries, your chemist had a model propose the candidate molecules. AI did real work on the way to the invention. So who is the inventor?

Notice that none of those inventions is itself an AI invention. “AI-assisted” refers to how the invention was made, not what it is. The rule is the same whether your AI-assisted invention is the latest model architecture or a new mousetrap.

Conception, which is the formation of a definite and permanent idea of the complete and operative invention, remains the touchstone of inventorship, and it is treated as a human act.

In February 2024, the USPTO issued its first Inventorship Guidance for AI-Assisted Inventions. It confirmed that AI-assisted inventions are not categorically unpatentable, so long as at least one natural person made a significant contribution to the claimed invention. To assess that contribution, the guidance asked examiners and applicants to apply the Pannu factors, a framework drawn from joint-inventorship case law.

In November 2025, the USPTO rescinded that February 2024 guidance in its entirety and issued Revised Inventorship Guidance for AI-Assisted Inventions.

The revised guidance withdraws the Pannu factors as a test for whether a single person using AI qualifies as an inventor and reinforces a simpler rule: the same legal standard for inventorship applies to every invention, whether or not AI was used. There is no separate or modified test for AI-assisted inventions. Where multiple people contribute to an AI-assisted invention, ordinary joint-inventorship principles, including the Pannu factors, still govern the analysis among the human contributors. Using AI tools does not change that analysis.

AI systems (including generative AI and other computational models) are treated as tools used by human inventors, comparable to laboratory equipment or specialized software. A tool does not qualify as an inventor. Only a natural person can be named under 35 U.S.C. § 100(f).

The natural-person rule extends to priority claims. A U.S. application cannot claim priority to a foreign application that names an AI system as an inventor, alone or jointly with humans.

AI-assisted is not the same as AI-generated

Note that an AI-assisted invention is not the same as a claim that an AI system itself invented something.

The guidance addresses the first situation: a person or team conceives an invention and uses AI as a tool along the way. That work is patentable in the ordinary course, provided a natural person is properly named.

A claim that no human conceived the invention runs directly into Thaler and cannot support a valid U.S. patent. Most real-world projects fall into the first category. But the line matters, because inventorship errors carry consequences. A patent can be held invalid or unenforceable if the named inventorship is incorrect and not properly corrected.

The practical message is consistent across both versions of the guidance: identify the natural persons who conceived the claimed invention, document their contribution, and do not name an AI system. Where AI tools play a meaningful role, internal records showing how the human team arrived at the invention can help support correct inventorship if it is ever questioned.

When AI is in the claims: they must still survive Section 101

The recurring obstacle for AI-related inventions is subject-matter eligibility under 35 U.S.C. § 101. Here the question that matters is whether you claim anything related to AI, not whether AI helped you invent it.

AI and machine-learning claims often draw abstract-idea rejections, because the USPTO and the courts treat mathematical concepts, mental processes, and certain methods of organizing human activity as judicial exceptions that are not, on their own, patent-eligible.

In July 2024, the USPTO issued the 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence. It applies the same two-step Alice/Mayo framework reflected in MPEP 2106 and adds three illustrative examples (Examples 47-49) aimed specifically at AI-related inventions:

  • Example 47: An artificial neural network used to detect anomalies, showing how a claim directed to a specific application of a neural network can be eligible by reflecting an improvement in computer technology.
  • Example 48: An AI-based method for separating individual speech from mixed audio (for example, multiple people speaking at once), showing how applying AI to a concrete technical problem in signal processing can be patent-eligible.
  • Example 49: Using an AI model to personalize medical treatment, showing how an AI model integrated into a particular practical application can be patent-eligible.

The thread that runs through these three examples is that an AI-related patent claim is more likely to be subject-matter eligible under Section 101 when it reflects a specific technical improvement or a practical application, not when it merely recites a mathematical model or instructs a computer to “apply” an abstract idea.

The two kinds of AI-related inventions noted above both go through this Section 101 analysis, but the path to patent-eligibility looks different for each.

Some AI-related inventions improve the AI technology itself. A new KV-caching scheme cuts the memory a model needs at inference. A quantization method lets a model run on cheaper hardware. A training pipeline converges in half the steps. Claims like these tend to have the best odds of surviving Section 101, because the invention is an improvement in how the computer itself works, the kind of technical improvement the guidance and the case law have long treated as patent-eligible.

Other AI-related inventions are products that use AI to do the product’s job. A mobile app uses a vision model to measure a room. A camera pipeline uses a model to deblur low-light photos. A battery controller uses a model to adjust charging as cells age. These can absolutely be patent-eligible, but the claim has to be anchored in the concrete technical problem the product solves, the way the claims in Examples 48 and 49 are. The danger zone is a business task with AI bolted on. A claim to using a model to decide which customers to email is, to an examiner, still a claim to deciding which customers to email.

In drafting, that points toward tying the AI to a concrete technical improvement, describing how the model interacts with hardware or a larger process, and explaining the technical problem the invention solves, rather than claiming the algorithm in the abstract.

Section 101 is not the only hurdle. AI-related inventions also raise written-description and enablement questions under 35 U.S.C. § 112, because a machine-learning model can be difficult to describe in a way that enables others to make and use the full scope claimed, alongside the usual novelty and obviousness questions under 35 U.S.C. §§ 102 and 103.

When AI helps prepare the paperwork: do you have to tell the USPTO?

Your engineers use AI to run prior-art searches and to write their invention disclosures. Your patent attorneys use AI to review those disclosures and to prepare the first draft of the patent application. At some point the question occurs to somebody: does the USPTO need to know?

The USPTO answered that question in April 2024, with guidance on the use of AI-based tools in practice before the Office.

The Office concluded that its existing rules are sufficient and did not create new AI-specific requirements. A few points matter for applicants and practitioners:

  • There is no general or per se duty to disclose that an AI tool was used (either in the inventive process or in preparing a filing) unless that use is material to patentability or the USPTO specifically requests it. The duty to disclose information material to patentability under 37 C.F.R. § 1.56 still applies and can extend to material AI use.
  • The duty of candor and good faith, and the duty of disclosure, rest on the individuals identified in the rules. They cannot be delegated to a computer or an AI tool.
  • A practitioner who uses AI to help draft claims must review and, where needed, modify those claims to present them in patentable form before filing. Every paper filed carries the signer’s certifications.
  • Existing obligations on confidentiality, foreign-filing licenses and export controls, and proper use of USPTO electronic systems all continue to apply when AI tools are involved.

What this means for your patent strategy

The United States uses a first-inventor-to-file system, which rewards getting a well-prepared application on file early. When AI shows up anywhere in your process, that means resolving inventorship, framing claims for § 101, and building out the specification before a public disclosure or a competing filing narrows your options.

Read together, the three guidance documents point toward a few concrete steps:

  1. Where AI helped your team invent: Name inventors carefully. Identify the natural persons who conceived the claimed invention and do not list an AI system. And document the human contribution, particularly where AI tools were used heavily; those records are your backstop if inventorship is ever questioned.
  2. Where AI is in the claims: Draft for Section 101. Frame claims around a specific technical improvement or practical application rather than an abstract model, and use the specification to explain the technical problem solved. And address Section 112 by describing the model and its training and operation in enough detail to support the full scope of what is claimed.
  3. Where AI helps with the paperwork: Treat AI as an aid, not a substitute for the practitioner’s review, certifications, and duty of disclosure.

How NK Patent Law approaches AI in the patent process

NK Patent Law is a patent prosecution and intellectual property strategy boutique that represents technology-driven companies from early development through commercialization. Because patent law is federal, our patent attorneys and patent agents represent clients regardless of location.

Our team has helped secure 2,100+ issued patents and 1,000+ registered trademarks for clients across 36 U.S. states and 27 countries, including computing-technology portfolios exceeding 100 patents for a single client.

Our managing partner, Doug Meier, is an electrical engineer by training who began his career as a software engineer on NASA’s Space Shuttle Program. Our attorneys and patent agents hold engineering and science degrees and have prosecuted patents in artificial intelligence and machine learning, wireless communications, digital image processing, and semiconductors — including hardware that accelerates AI workloads.

That technical fluency matters when an examiner’s rejection turns on how a neural network actually improves a computer’s functioning, when an inventorship question depends on understanding what the human team contributed, or when AI drafting tools need to be used with the review and judgment the USPTO expects of the practitioner signing the papers.

NK Patent Law has been recognized by Legal 500 U.S. Elite, Best Lawyers in America for Intellectual Property, Business North Carolina Legal Elite, and the Chambers USA Regional Spotlight Guide.

Speak with a patent attorney

If AI shows up anywhere in your patent process, in the inventing, in the invention, or in the filing, a well-planned application can help you protect what matters and avoid costly surprises later.

Call NK Patent Law at (919) 348-2194 or contact us online to discuss your patent strategy.