When you bring an AI, software, or business-method (BM) invention to Korea, the first question is whether the claim qualifies as a patentable invention at all. Korean Patent Act Article 2 defines an invention as "a highly advanced creation of technical ideas utilizing a law of nature," and KIPO's BM patent guide is explicit: a pure business method on its own is not patentable. The claim has to be combined with hardware — computer, network, processor — to clear the eligibility bar.
What is patent-eligible in Korea?
Article 2 defines an invention as "a highly advanced creation of technical ideas utilizing a law of nature." Pure AI algorithms or business procedures are products of human reasoning, contracts, and conventions — they don't "utilize a law of nature" in themselves and so are not patent-eligible as such. Combined with hardware-coupled information processing, however, they cross into patent-eligible territory.
KIPO's computer-related invention guidelines require that information processing by software on a computer is concretely realized using hardware. So a bare "method comprising the steps of …" claim is risky — preferred forms are "a system in which the processor performs …" or "a non-transitory computer-readable medium having a program recorded thereon that causes a computer to perform …."
| Claim form | Eligibility | Example |
|---|---|---|
| Pure business method | Not eligible | "Method of granting discounts based on customer tier" |
| Pure algorithm / math | Not eligible | "Method of multiplying input matrices" |
| HW-coupled system | Eligible | "A system comprising a processor and memory configured to …" |
| Program-on-medium | Eligible | "A non-transitory computer-readable medium storing a program that causes a computer to …" |
| Method (with HW actor) | Eligible | "A method, performed by at least one processor, comprising: …" |
Four Korean claim patterns that work
- System claim: "A system comprising at least one processor and memory, the processor configured to …" — naming hardware in the preamble
- Method claim with HW actor: "A method performed by at least one processor, comprising: …" — name the actor ("the processor", "the server") in each step
- Program / medium claim: "A non-transitory computer-readable medium storing a program that causes a computer to perform …"
- Model / data-structure claim (AI): "A neural-network model trained on … and configured to output … from input …" — describe the training procedure and data layout in the spec
US §101 — the Alice/Mayo two-step
The US uses a different framework. The USPTO Subject Matter Eligibility guide describes Alice/Mayo: (1) is the claim directed to an abstract idea, law of nature, or natural phenomenon; and (2) if so, does the claim as a whole integrate the exception into a practical application. Tacking on "performed on a computer" is not enough.
The July 2024 guidance update added new AI-specific examples (Examples 47-49): a neural network for anomaly detection, AI-based speech separation, and AI-driven personalized medicine. Each is recognized as either an "improvement to the functioning of a computer" or an "improvement to another technical field." An August 2025 memo further raised the bar for §101 rejections — overall a friendlier environment for AI/SW filings.
Korea vs US — strategic differences
| Aspect | Korea (KIPO) | US (USPTO §101) |
|---|---|---|
| Test | Invention definition (natural law + tech idea) | Alice/Mayo two-step |
| Core requirement | HW coupling, concrete realization | Practical application |
| Common failure | Pure business method / algorithm | Abstract idea on a computer |
| AI-specific guidance | Computer-related invention guidelines | 2024 update + Examples 47-49 |
| Claim formats | System / method / medium all fine | Same forms, but drafting is fussier |
AI claim — a Korean-style example
Korean-style AI claim: "An object-detection system comprising at least one processor and memory, wherein the memory stores convolutional-neural-network weights trained to detect objects in an input image, and wherein the processor is configured to (1) receive the input image, (2) compute candidate object regions based on the weights, and (3) classify the candidate regions." Naming the system, memory, processor, and step-by-step operations together is what gets the claim through Korean BM/SW examination.
Frequently asked questions
Is a pure-algorithm claim absolutely unpatentable in Korea?
Yes — a claim that is just "a mathematical algorithm" with no hardware actor is rejected. Reframing the same invention as "a system that performs the algorithm" or "a medium storing the algorithm" can clear the bar. The form, not the underlying invention, is the issue.
I got a §101 rejection in the US — will Korea grant the same claim?
Possibly. Alice/Mayo can be stricter than KIPO's BM/SW guidelines for some claim shapes, and the same invention written in Korean-friendly form (HW coupling, concrete information processing) may be allowed in Korea. Plan to draft the Korean and US claim sets separately rather than translating one for the other.
Can AI training data itself be patent-protected?
Pure data is generally not eligible, but a system that combines a specific data structure with processing methods can be claimed. For example: "a method of training a neural network using training data structured as …, and a system performing the method." Beyond patents, treating the dataset itself as a trade secret is also worth evaluating.
File AI / software / BM patents with iphere
We draft Korea-friendly HW-coupled claims and a US §101-aware narrative in the same specification.