Founding Partner
AI patent eligibility under Section 101 is determined by a two-step framework that asks whether your claims are directed to an abstract idea and, if so, whether they add something more. The 2024 USPTO guidance gives applicants specific examples of what passes and what does not. Understanding that guidance before you file is one of the most practical things you can do to protect your AI invention.
Richard Gearhart, founding partner of Gearhart Law, has spent nearly 30 years helping technology companies navigate the most complex areas of patent law. Section 101 rejections for AI inventions are one of the most common and most misunderstood obstacles applicants face today.
This post explains exactly how the eligibility framework applies to AI, what the current USPTO guidance says, and what it means for how you structure your application.
What Section 101 Actually Says
Section 101 of the Patent Act states that anyone who invents or discovers any new and useful process, machine, manufacture, or composition of matter may obtain a patent. On its face, that covers almost everything. The problem is that courts have carved out three judicial exceptions: abstract ideas, laws of nature, and natural phenomena.
For AI inventions, the abstract idea exception is the one that matters. Machine learning models, neural networks, and data processing methods all involve mathematical operations. To an examiner applying the Section 101 framework, an AI claim that describes inputs being processed to produce outputs can look indistinguishable from a mathematical formula applied to a general-purpose computer.
The Two-Step Framework: What Examiners Actually Do
When an examiner reviews an AI patent application under Section 101, they follow a structured analysis with two main steps.
Step 1: Is the claim directed to a judicial exception?
The examiner first looks at what the claim is actually about. If the claim is directed to a mathematical concept (a formula, an algorithm, or a mathematical relationship), it is directed to a judicial exception and proceeds to step two. Most AI claims trigger this step because training a model, running inference, or optimizing weights are all mathematical operations at their core.
Step 2A: Does the claim do enough to survive?
Step 2A involves two separate questions the examiner works through in order.
Prong 1 – Practical application: Does the claim integrate the mathematical concept into a real-world technical context? A claim that uses a neural network to improve the efficiency of a specific industrial process may pass here because the math is applied to a concrete technical problem with a defined real-world benefit.
Prong 2 – Something more: If the claim does not clearly integrate the exception into a practical application, does it add something significantly more, like an inventive concept that goes beyond the abstract idea itself?
Two things that do not pass either prong:
- Reciting generic computer components like a processor or a server
- Describing the inventive concept only in the specification, not in the claims themselves
Step 2B: Is there an inventive concept beyond the abstract idea?
If a claim fails both questions in Step 2A, the examiner asks one final question: do the additional elements in the claim amount to significantly more than the abstract idea, law of nature, or natural phenomenon at the core of it?
Routine and conventional activities, like gathering data, storing results, and displaying output, do not pass this test. A genuinely novel technical implementation might.
The USPTO AI Guidance: What It Says and Why It Matters
In 2024, the USPTO issued updated guidance specifically addressing AI-assisted inventions and patent eligibility. This guidance does not change the law. What it does is give examiners and applicants specific examples of how to apply the Alice framework to AI inventions.
What the guidance confirms
- A claim to training a neural network is not automatically abstract. It depends on what the claim says about how the training is done and what technical improvement it produces.
- A claim to an AI method tied to a specific practical application, improving the accuracy of a defined diagnostic process, reducing latency in a specific type of network communication, or optimizing a manufacturing process with measurable efficiency gains is more likely to survive Step 2A, Prong 1.
- Generic references to using AI or machine learning without specifying the technical implementation do not satisfy the practical application requirement.
What the guidance warns against:
- Claims that describe the result of applying AI without describing the specific technical means of achieving it.
- Claims where the only practical application described is the mathematical result itself, for example, producing a more accurate prediction without a defined real-world technical context.
- Claims that rely on post-solution activity, such as displaying a result or storing an output, to satisfy the practical application requirement, without integrating the AI into the technical process itself.
Gearhart Law helps AI companies and inventors structure patent applications that address Section 101 from the start, before a rejection forces a difficult amendment. Reach out for a free half-hour consultation.
AI-Assisted Inventions vs. AI-Generated Inventions
Only natural persons can be listed as inventors on a U.S. patent application. An AI system cannot be an inventor or co-inventor regardless of how much it contributed, confirmed by the Federal Circuit in Thaler v. Vidal (2022).
| AI-Assisted | AI-Generated | |
| Human contribution | Human evaluated, selected, and refined AI outputs | Little or no meaningful human creative input |
| Inventorship | Human inventor | Inventorship problems, no valid human inventor |
| Section 101 risk | Manageable with the right claim structure | Compounds existing eligibility issues |
| Patentable? | Yes, if claims show a concrete technical improvement | Significant legal obstacles |
For AI-assisted inventions, the claim needs to show not just that AI was used, but that the specific way it was used produces a concrete technical improvement. That is what separates a patentable invention from an abstract computational result.
How Section 101 Rejections Play Out in AI Prosecution
Understanding the framework is one thing. Knowing what happens when an examiner applies it incorrectly or applies it in a way that is technically correct but can be rebutted is where the real prosecution strategy lives.
When examiners get it wrong
Examiners sometimes issue Section 101 rejections for AI claims that integrate the mathematical concept into a specific practical application, either because the practical application is not made explicit enough in the claim language or because the specification does not make the technical improvement concrete.
These rejections can be successfully rebutted by amending the claims to make the practical application explicit and by pointing to specific passages in the specification that demonstrate the concrete technical benefit.
When the rejection is harder to overcome
A Section 101 rejection is much harder to overcome when the claims genuinely are directed to a result rather than an implementation. Adding reference to a processor or a server does not help.
The fix requires amending the claims to describe the specific technical means of achieving the result, which often requires evaluating whether the specification contains enough technical detail to support narrower, more specific claims. If it does not, the amendment options are limited, and a continuation strategy becomes important.
The role of the examiner interview
For Section 101 rejections specifically, an examiner interview is often the most efficient path to resolution. Examiners handling AI applications deal with complex technical subject matter, and a well-prepared interview that clearly explains the specific technical improvement and ties it to the current USPTO guidance can resolve a rejection that would otherwise require multiple rounds of written responses.
Build Your AI Patent Application Around Eligibility From the Start
Section 101 is not an afterthought in AI patent prosecution. It is the first question every examiner asks and the most common reason AI applications get rejected. The applicants who navigate it successfully are the ones who understood the framework before they filed and built their claims and specification around it from the beginning.
Gearhart Law helps AI companies throughout New Jersey, New York, and beyond build patent applications that are structured to survive Section 101 scrutiny. Leave your details, and we will be in touch, or call 908.273.0700 for a free half-hour consultation.
Frequently Asked Questions About AI Patent Eligibility Under Section 101
1. What is Section 101, and why does it matter for AI patents?
Section 101 is the statute that defines what kinds of inventions can be patented in the United States. For AI inventions, it matters because courts have carved out an exception for abstract ideas, and mathematical operations, which underlie most AI systems, can look abstract to a patent examiner.
2. What is the Alice framework, and how does it apply to AI?
The Alice framework comes from the 2014 Supreme Court decision in Alice Corp. v. CLS Bank. It established a two-step test for AI patent eligibility: first, is the claim directed to an abstract idea? Second, does it add something significantly more? For AI, step one is almost always triggered because machine learning involves mathematical operations. The real question is step two: does the claim connect the math to a specific real-world use, or does it add something genuinely new that goes beyond the math itself?
3. What did the 2024 USPTO AI guidance change?
The 2024 USPTO guidance gave examiners and applicants specific examples of how the Alice framework applies to AI inventions, with particular focus on when AI claims are tied to a practical application versus when they are directed to a result of applying math. The guidance confirms that training a neural network is not automatically abstract and that claims tied to specific technical improvements in a defined real-world context are more likely to survive examination.
4. Can I patent an AI invention if AI helped create it?
Yes, as long as a human inventor made a significant contribution to the conception of the claimed invention. The USPTO and the Federal Circuit have confirmed that only natural persons can be listed as inventors. If you used AI tools to assist in development, generating options you then selected and refined, you are the inventor. If the AI generated the invention autonomously without meaningful human creative contribution, the application faces inventorship problems in addition to any Section 101 issues.
5. What makes an AI claim pass the practical application test?
A claim passes the practical application test when it integrates a mathematical concept into a specific real-world technical context that produces a concrete, measurable benefit. Examples include an AI method that improves the efficiency of a specific industrial process, reduces error rates in a defined diagnostic application, or optimizes network performance in a measurable way. Generic references to using AI or machine learning to produce better results, without specifying the technical context or the means of achieving the improvement, do not satisfy the test.
6. What should I do if I receive a Section 101 rejection on my AI application?
Evaluate whether the practical application is explicit enough in the claim language and whether the specification contains enough technical detail to support an amendment. An examiner interview is often the most efficient path to resolution for Section 101 rejections, as it lets you explain the technical improvement directly before committing to a written amendment.
7. Is it better to address Section 101 before filing or after a rejection?
Before filing, without question. Amending claims after a Section 101 rejection is possible but limits your options, especially if the specification does not contain enough detail to support the amendments you need. Reviewing your claims against the current USPTO AI guidance before filing, making the practical application explicit in both the claims and the specification, and getting a legal patentability opinion from an experienced AI patent attorney all cost less than a protracted prosecution with multiple Office Actions.