The Secret to Patenting AI/ML inventions in Singapore: A Guide for Innovators

4 August 2025

For an inventor, engineer, or investor, navigating patent law can feel like learning a new language. 

On 11 October 2024, the Intellectual Property Office of Singapore (“IPOS”) released a Supplemental Guidance (“Guidance”) for Examination of AI-related Patent Applications. The Guidance is intended to be a quick patentability reference for applicants seeking to protect their artificial intelligence (“AI”) related inventions. 

The guidance on AI inventions, however, contains a clear and consistent message. The “secret” to getting your AI invention patented isn’t a legal trick; it’s about how you frame your invention.

In short, the secret is: Your AI must be presented as a practical solution to a specific, technical problem.

You cannot patent a pure algorithm or a mathematical formula in the abstract. Instead, you must show how your AI is applied in the real world to achieve a technical effect. Think of it as the difference between patenting the concept of addition and patenting a physical calculator that uses addition to solve engineering problems.

IPOS uses a three-step test to decide this, but for you, the innovator, it boils down to one key question: What is my invention’s “actual contribution” to the world, and is that contribution purely abstract?

Here’s how to ensure your contribution is seen as concrete and technical, based directly on the IPOS guidelines.

1. Anchor Your Invention to a Specific Problem

Your patent application must go beyond describing what your AI is (e.g., “a neural network”) and focus on what it does to solve a well-defined problem.

• Generic (and likely to fail): “A method for controlling a system using AI.” This is too broad. It could apply to anything from a toaster to a power grid without explaining the specifics.
• Specific (and likely to succeed): “A method for computing nozzle blockage in a liquid metal 3D printer by analyzing captured images of the nozzle” (Example 9). This is a highly specific, real-world industrial problem.

Actionable Advice: Clearly state the specific, practical problem you are solving in your patent application. Describe the context and the pain points. For example, instead of saying your AI “optimizes a process,” explain that it “predicts the tensile strength of a steel sheet by analyzing specific line production factors, environmental factors, and material component values”.

2. Ensure the Problem is Technical, Not Just Business-Related

Even a specific problem might not be patentable if it falls into an excluded category, like a “method of doing business.” This is one of the most critical hurdles for AI patents.

• The Stock Trading Trap (Example 2): The document gives an example of an AI that creates a strategy for executing a large stock trade to minimize price impact. This is a very specific problem. However, IPOS considers a stock trading strategy to be a “pure business method.” Since the ultimate goal is financial and not technical, it’s not patentable, even with clever AI.
• The Fraud Detection Success (Example 1): In contrast, an AI system that uses a Generative Adversarial Network (GAN) to detect if an ID document is fraudulent is patentable. While this has business implications, the core task is technical: analyzing image data against a template to verify authenticity.

Actionable Advice: Frame your invention’s purpose in technical terms. Ask yourself: Does my AI solve a problem in a field of science or engineering? Does it improve a manufacturing process, a diagnostic tool, or the functioning of a machine? If the end goal is purely economic, administrative, or commercial, you risk rejection.

3. Detail the “How”: Connect Your Inputs and Outputs to the Technical Task

It’s not enough to say your AI solves a problem. You must describe how it’s functionally tied to that problem. This means detailing the specific data you use and the specific output it produces.

• Vague (and likely to fail): “A method to train a neural network by adding new nodes.” This describes a general technique for improving any AI model but isn’t tied to a specific application.
• Detailed (and likely to succeed): “A method of predicting a cardiac event by training an AI with specific inputs: heart rate variability from an ECG, systolic blood pressure, and pulse oximetry”. Here, the inputs are explicitly linked to the technical goal of medical diagnosis.

Actionable Advice: Your patent application should read like a technical specification, not a marketing document.

• Specify Your Inputs: What exact data does your AI need to function? (e.g., “gas species fractions and a cooling gas injection volume”).
• Specify Your Outputs: What is the direct, tangible result? (e.g., “the tensile strength of the steel sheet” from Example 7 or “a blockage fraction of the nozzle”).

4. When the Invention is the Computer Process Itself

Sometimes, the invention isn’t about an external task but about making the computer or the AI itself run better. This is also considered a technical contribution.

• Optimizing the AI for the Hardware: An AI that reduces its own size by removing less important nodes to match the number of calculators in a specific processor is patentable. The technical problem is improving computational efficiency and speed.
• Optimizing Memory Usage: A method that optimizes how a neural network uses a computer’s volatile memory by defining where to store temporary data is also patentable. The problem it solves is technical: more efficient use of computer resources.

Actionable Advice: If your invention is about improving the performance of an AI model or the computer running it, focus on those technical benefits. Explain how your method leads to faster processing, lower memory consumption, or more efficient use of hardware.

Final Checklist for Your AI Invention

When writing the invention disclosure for your invention, ask yourself these questions:

1. What is the one specific problem my AI solves? Can I describe it in a single, clear sentence?
2. Is this problem technical in nature? (e.g., related to manufacturing, engineering, diagnostics, data processing) Or is it purely a business, financial, or administrative method?
3. Have I clearly described how my AI solves the problem? Have I detailed the specific inputs it uses and the tangible, technical output it produces?
4. If my invention improves the AI itself, have I explained the technical benefit? (e.g., faster speed, less memory).

By focusing on these principles from the IPOS guidance, you can frame your AI invention not as an abstract algorithm, but as a tangible, technical tool that solves a real-world problem—and that is the secret to clearing the patent eligibility hurdle.

Our Expertise

Auriga IP has partnered with leading companies and innovative startups in Singapore, guiding them through their most critical and complex intellectual property challenges. Our expertise lies in leveraging patents, trademarks, and registered designs to protect creativity and in developing strategic IP roadmaps to establish market dominance. We are dedicated to delivering unparalleled service, effective solutions, and expert advice for all your IP needs. We invite you to contact us for a complimentary consultation on how we can help secure your intellectual property assets.

Disclaimer: Please note that the content on aurigaip.com, including pages, articles, and comments, is not legal advice and does not establish an attorney-client relationship. The views expressed in published articles are solely those of the author at the time of publication and do not necessarily reflect the opinions of the author’s employer, clients, or the sponsors of aurigaip.com