Patent Eligibility for Software and AI Inventions: Section 101 Basics

How the Alice framework applies to software and AI inventions, what examiners look for under Section 101, and how drafting choices affect eligibility.

By , registered patent attorney. Published . 4 min read.

General information about patent law and practice, not legal advice about your situation. How these rules apply depends on the facts.

Software and AI inventions can be patented in the United States, but they face a hurdle that many other technologies do not: the patent eligibility requirement of 35 U.S.C. 101. Claims that courts or examiners view as directed to an abstract idea, such as a mathematical calculation, a mental process, or a way of organizing human activity, are not eligible unless they add something more. How a software invention is described and claimed has a large effect on how that question comes out.

The Alice framework

The governing test comes from the Supreme Court’s decisions in Mayo Collaborative Services v. Prometheus Laboratories (2012) and Alice Corp. v. CLS Bank International (2014). It has two steps:

  1. Is the claim directed to a judicial exception, such as an abstract idea?
  2. If so, does the claim include additional elements that transform it into a patent-eligible application of that idea, sometimes called an “inventive concept”?

In Alice, claims to a computerized method of mitigating settlement risk were held ineligible because implementing an abstract financial idea on a generic computer was not enough.

How the USPTO applies the test

Examiners follow the USPTO’s eligibility guidance in MPEP 2106, which breaks the analysis into steps:

  • Step 1: Is the claim to a process, machine, manufacture, or composition of matter?
  • Step 2A, prong one: Does the claim recite a judicial exception? For software, that usually means an abstract idea in one of the recognized groupings: mathematical concepts, mental processes, or certain methods of organizing human activity.
  • Step 2A, prong two: If so, does the claim as a whole integrate the exception into a practical application, for example by reflecting an improvement in the functioning of a computer or another technology?
  • Step 2B: If not, do the additional elements amount to significantly more than the exception, beyond well-understood, routine, and conventional activity?

A claim that is not directed to a judicial exception, or that integrates one into a practical application, is eligible without reaching Step 2B. The practical-application question is often where software and AI claims are won or lost.

Recent USPTO guidance has emphasized that analysis. An August 2025 memorandum reminded examiners to evaluate claims as a whole at that step and to reserve rejections for cases where ineligibility is more likely than not. In November 2025, the Director designated as precedential the Appeals Review Panel decision in Ex parte Desjardins, which found that claims reflecting an improvement in how a machine learning model operates could integrate an abstract idea into a practical application. Guidance in this area continues to develop, so the current versions are worth checking.

What tends to help

Federal Circuit decisions illustrate the kinds of claims that have been found eligible. In Enfish, LLC v. Microsoft Corp. (2016), claims to a self-referential database table were eligible because they were directed to an improvement in how computers store and retrieve data. In McRO, Inc. v. Bandai Namco Games America Inc. (2016), claims to specific rules for automating lip-synchronized animation were eligible because they recited a particular way of achieving a result rather than the result itself.

The key distinction is between an improvement to computer functionality or another technology, and simply using a computer as a tool to carry out a task faster or at scale. Specific technical detail helps show the first, but detail alone does not make a claim eligible: a long list of conventional steps applying a mathematical idea can still be ineligible. Drafting approaches that draw on those principles include:

  • Describing the technical problem and the technical solution. Explain what was wrong with existing systems and how the invention changes how the computer, network, or model operates.
  • Claiming the specific way, not just the result. A claim to “determining a risk score” says what is achieved. A claim reciting the particular data structures, model architecture, training process, or sequence of operations that produce the improvement says how, and is more likely to reflect a technical solution, as long as those features are what actually deliver it.
  • Tying the steps to concrete effects. Reduced latency, lower memory use, improved accuracy on a defined task, or control of a physical device are more persuasive when the description explains why they follow from the claimed steps.
  • Supporting the improvement in the specification. An improvement asserted only in a response, without support in the application, carries less weight.

AI inventions in particular

AI inventions raise the same questions in a particular form. The USPTO’s July 2024 eligibility update added AI-specific examples (Examples 47 to 49) that illustrate the line. Claims that simply apply a known machine learning technique to a new field of data, or that recite training and using a model in general terms, are often characterized as abstract. Claims directed to a specific improvement in how a model is built, trained, or deployed, or that use its output in a particular technical application, are generally on firmer ground. Federal Circuit decisions remain the controlling authority on eligibility, and they do not always track USPTO guidance, so a claim that satisfies the examiner can still face scrutiny later.

Responding to a Section 101 rejection

Common response strategies include explaining, with citations to the specification, how the claims improve the functioning of a computer or another technology; amending the claims to recite the specific technical features that achieve that improvement; and, where appropriate, discussing the rejection with the examiner in an interview. Responses are most effective when the original application already contains the technical detail the argument needs.

A practical takeaway

For software and AI inventions, eligibility is best addressed when the application is drafted, not after a rejection. When preparing an invention disclosure, include the technical problem, why prior approaches fell short, and the specific mechanism that solves it. Our software invention disclosure checklist covers the details worth capturing.