The growth of Generative AI (“GenAI”) technologies has accelerated over the past few years. GenAI, which is able to create new content by learning from existing data, is reshaping entire industries. The release of OpenAI’s ChatGPT in November 2022 exposed the public to the vast real-world applications of GenAI.

The rapid growth of GenAI has also led to efforts to increase patent protections. China and the United States are the current leaders in GenAI technology innovation and GenAI patenting. Some of the companies with the most GenAI patents include Tencent, Ping An Insurance Group, Baidu, and IBM.

Companies must be vigilant in seeking patent protections to safeguard their GenAI innovations. The main types of data used in GenAI patents include speech, video, images, music, and sound. While patent offices around the world have issued guidance about the patentability of certain subject matter relating to AI, the legal framework is not concrete.

The 2014 Supreme Court case Alice Corp. v. CLS Bank International established a two-part test for determining whether software-related applications are patentable subject matter. The first step involves determining whether a claim encompasses an abstract idea or other patent ineligible concept. If so, the next step involves considering whether there are additional elements that would transform the claim into a patent-eligible application.

The decision in the Alice Corp. v. CLS Bank International case has broad implications for assessing whether certain GenAI applications are patentable. In 2024, the U.S. Patent and Trademark Office (USPTO) issued its “Guidance Update on Patent Subject Matter Eligibility”, which included examples pertaining to AI-related applications. The guidance suggests that the patent eligibility standards may be satisfied for inventions that involve training an AI model in a way that results in noticeable improvements or practical applications.

Patent offices in other prominent jurisdictions have released similar guidance with respect to the patentability of GenAI inventions. The European Patent Office (EPO) has indicated that inventions involving AI must have a “technical character.” The Japan Patent Office (JPO) has provided GenAI case examples to illustrate the patent requirements in different scenarios.

There are three main types of GenAI models: generative adversarial networks (GANs), variational autoencoders (VAEs), and decoder-based large language models (LLMs). Most patent filings related to GenAI are currently in the GANs category. A GAN is a deep learning model consisting of two neural networks in competition with each other: a generator and a discriminator. When presented with a training data set, the “generator” produces fake data and the “discriminator” learns to detect the fake data and distinguish it from real data. GANs are often associated with tasks that involve distinguishing images. Although many patents belong to the GANs category, there has been significant growth in patents for LLMs. LLMs are associated with natural language processing tasks and form the basis of many chatbots.

While patents offer companies with a certain degree of protection, it is possible for competitors to design around patented AI technologies. Since patents require some public disclosure, competitors can potentially use the available information to modify algorithms and achieve similar outcomes. Nonetheless, companies should understand the evolving patent landscape and apply for GenAI patent protection as appropriate.