Patenting AI Models: Avoiding the Dreaded Subject Matter Objection

October 31, 2023

Insights Intellectual Property

In celebration of Small Business Month, Vector in collaboration with Smart & Biggar LLP, developed a series exploring the dynamic relationship between AI and IP. In this series, designed to help support Canadian startups and research professionals, Vector dives into how AI innovation intersects with intellectual property, discussing trends, challenges, and strategies that shape this ever-evolving landscape.

When you try to patent your AI model, you should be wary of a nefarious legal objection: patentable subject matter. Patentable subject matter is an area of law that has a long history of being unclear and highly contested. Fortunately, the inventive aspects and technical character of a new AI model are often relatively straightforward; they likely have a readily identifiable new additional element or novel model structure. As such, the courts and examiners have been relatively kind to inventive AI models, as long as the invention has a stated purpose.

Technical Character and Purpose

In the realm of software, patents face specific challenges, especially in jurisdictions like Canada and several others, where software may encounter statutory subject matter objections frequently. Having software as part of an invention neither enhances nor diminishes its patentability in principle. However, if a generic computer is used to automate or digitize a known process, securing patent protection can be difficult, even with substantial effort.

To overcome these challenges, it is essential to find aspects of the technology with a technical character, such as achieving technical effects, solving technical problems, or having well-defined technical features in the claims.

An easy conceptualization for technical character is the “Does an engineer think it’s cool?” test. If an engineer in the field finds an invention impressive, it is more likely to pass the subject matter test and be patentable.

On the other hand, if the engineer considers it standard or known with an application to an existing problem domain, it may not meet the requirements for patent protection, both from a technical character perspective and other requirements (novelty and obviousness). The standard for patent protection is based on a person “skilled in the art” without having the benefit of hindsight analysis, often characterized as a skilled technician. As such, while you consider an invention to be obvious based on your expertise, you may be basing your conclusion on a much higher standard than required by the patent process.

The ideal description for an AI invention should describe the technical problem solved by the inventors, which can be characterized as a “computer problem” (e.g., use of AI vs. conventional, non-AI implementations). This “technical purpose” is needed in some jurisdictions for the invention not to be deemed as directed towards an abstract mathematical idea. It is also typically wise to establish that the hardware or software is essential by providing a significant amount of technical detail regarding the hardware and software used to solve the technical problem.

Want more?

In the world of AI and machine learning, IP is a critical asset. Stay ahead of the competition by making IP an integral part of your business strategy.

Example Claims for AI Models

In many cases, inventors bring forward new and innovative model structures for machine learning. These models often offer advantages in terms of improved performance during both training and inference. For instance, a patent claim could be directed to a novel model structure, such as a Recurrent Neural Network (RNN) with an additional element called an Auxiliary Sentinel Gate (see Figure 1).

Figure 1: An example claim for a new model (Sentinel LSTM US20180144248A1)
Figure 1: An example claim for a new model (Sentinel LSTM US20180144248A1)

This gate plays a crucial role in modulating the stored auxiliary information from the memory cell for the next prediction, thereby enhancing the recurrent functionality of the RNN.

Another example of potential patentable subject matter could involve a novel AI model designed for audio feature extraction. This model incorporates convolutional neural network layers specialized for audio processing, showcasing a unique structure tailored for a specific application.

These examples demonstrate the types of inventions that can be protected through patent claims. When inventors introduce new model structures that deliver improved performance or offer specialized capabilities, they may have the opportunity to secure patent protection for their innovations in the field of machine learning.


To claim a novel AI model, it is important that you consider and include a purpose to the invention as to avoid directing your invention towards an abstract mathematical concept. If you, as an AI inventor/engineer/developer, believe that the invention is “cool”, it will likely pass the technical character requirement. Patenting AI inventions can be a tricky process; it is usually advisable to retain professional assistance in order to have a good, valuable patent.

Explore other articles in this series

Three people stare at a laptop with a Vector logo on it

Intellectual Property and Generative AI: Many Questions, Few Answers

Ownership of IP: Do You Own Your Invention?

Two women typing code onto a computer

Patent Searching: How to Find Out if You Really Invented Something

A man looks at a white board with red formulas on it

Protecting inventions related to improvements to data and novel inputs and outputs

Download our white paper on IP for AI in Canada!

IP is a key business asset, especially in AI and machine learning. To gain a competitive edge, prioritize integrating IP into your business plans, including strategies for commercialization and monetization.


Vector Faculty Member Gautam Kamath

Vector researcher Gautam Kamath breaks down the latest developments in robustness and privacy

Large Language Models

Standardized protocols are key to the responsible deployment of language models

A man looks at a white board with red formulas on it
Trustworthy AI

How to safely implement AI systems