BitLaw

AI Inventions are Patentable

The United States has long recognized the patentability of inventions embodying artificial intelligence (AI). Some of the earliest AI patents include U.S. Patent No. 3,308,441 (filed in 1963), relating to an artificial neural network, and U.S. Patent No. 3,287,649 (also filed in 1963), relating to an audio signal pattern recognition using a neural network.

The U.S. Patent and Trademark Office (USPTO) grants patents for AI-based inventions as long as they meet the requisite criteria of novelty, non-obviousness, utility, and patent eligibility (described below). Whether it is a novel machine learning algorithm, a unique AI hardware component, or a groundbreaking application of neural networks, the U.S. patent system provides a means for inventors to secure the exclusive rights to their AI-driven inventions.

While AI can be a powerful tool in the development of inventions, U.S. law currently states that only natural persons can be named as inventors on U.S. patents. Mr. Stephen Thaler tested this policy by filing for patent protection on an invention that was invented solely by an artificial intelligence system. The USPTO rejected the application, insisting that only human inventors can be granted patent protection. The Federal Circuit upheld this decision (see Thaler v. Vidal). However, that does not mean that artificial intelligence can not participate in the inventive process with a human.

Technologies Where AI Patents Have Been Granted

Artificial Intelligence (AI) has ushered in a new era of innovation, and not surprisingly, the United States Patent and Trademark Office (USPTO) has granted hundreds of thousands of patents on this technology. Generative AI in particular has seen significant patent activity recently. This includes patents on Large Language Models (LLMs) that generate human-like text, and other media trained models that are capable of generating images, audio, and video. Patents in this area cover innovations in training methods and model architectures. For instance, patents are being issued for methods that improve the efficiency and scalability of LLMs, as well as for systems that enhance the quality and realism of generated content.

Patents can also be filed on the useful application of these technologies, such as automated content creation, realistic image synthesis, and sophisticated natural language processing. Multi-modal models, which integrate both input and output capabilities across text, images, and audio, are also a key focus, with patents protecting techniques that allow these models to understand and generate multi-faceted content.

Generative Adversarial Networks (GANs) are another major area of patent activity. GANs consist of two neural networks, a generator and a discriminator, that work together to produce realistic data. The generator creates new data instances, while the discriminator evaluates them for authenticity. This adversarial process results in the generation of highly realistic images, videos, and audio. Patents related to GANs cover a variety of innovations, including novel architectures and training techniques that enhance the stability and performance of these models. For example, there are patents on methods for reducing the common issues of GANs, such as mode collapse and training instability. Additionally, patents have been granted for specific applications of GANs in fields like medical imaging, where they are used to generate high-quality synthetic images for training other AI models, and in the entertainment industry, where they help create realistic visual effects and animations.

Another noteworthy AI type is Reinforcement Learning (RL), which is centered around an agent learning to make decisions by interacting with an environment. Through a system of rewards and penalties, the agent learns optimal strategies for decision-making, a principle that has been employed in various fields including robotics, game playing, and autonomous systems. Patent activity in RL revolves around novel algorithms and systems that enable machines to learn complex behaviors and make intelligent decisions over time.

Moreover, Neural Symbolic Computing represents a fusion of neural networks with symbolic reasoning, aiming to harness the strengths of both machine learning and logic-based AI. This hybrid approach enables the learning of structured representations from data and the reasoning over these representations in a symbolic manner. Patents in this realm may cover novel architectures, learning algorithms, and applications that meld symbolic reasoning with sub-symbolic learning.

These emerging AI technologies join the ranks of established fields like Knowledge Processing, Speech Recognition, and AI Hardware, among others. Knowledge Processing encapsulates methods to represent facts about the world and derive new knowledge from a database, forming the foundation of expert systems. Similarly, AI Hardware, exemplified by innovations like Google's Tensor Processing Unit (TPU) encompasses the physical components necessary to implement AI software, marking significant strides in enhancing the efficiency of running neural network algorithms.

Introduction to Patent Eligibility Under Section 101

As is true for other software inventions, patent applications covering artificial intelligence must be eligible for patent protection under Section 101 of the Patent Act before a patent will be granted. In particular, the application must be considered "patent eligible" under the guidelines from the Supreme Court’s Alice Corporation decision (2014). Determining the eligibility of a particular AI invention for patent protection is, unfortunately, a complicated issue. While Section 101 forms the basis for this requirement, the patent statute does not describe the issue at all. Instead, courts have developed the law and test for patent eligibility, which makes interpreting this issue more difficult.

The Bitlaw website contains a great deal of information on this topic, which can be found through the Section 101 Index. The considerations for AI-related inventions’ subject matter eligibility are largely similar to those for other software-related inventions. Bitlaw provides Section 101 Guidance on how to analyze these issues, including examples on how an attorney might handle a Section 101 rejection. In addition, the USPTO has issued some specific guidance to cover artificial intelligence inventions, including hypothetical Example 39, "Method for Training a Neural Network for Facial Detection." A separate page has been created in Bitlaw to discuss how this example has been treated by the Patent Office, and to provide additional considerations of how Section 101 applies to AI-related inventions.

Can AI Invent a Patentable Invention (AI Inventorship)?

In 2022, the Federal Circuit held in Thaler v. Vidal that only natural persons can be inventors under the U.S. Patent Act, which means that any invention created entirely by an AI system is not patentable in the United States (see Bitlaw’s discussion of Thaler). The decision, however, did not address the far more common situation in which AI assists a human inventor, leaving uncertainty about how to evaluate inventions that arise from a collaboration between human ingenuity and AI-generated contributions.

To address this gap, the USPTO issued its 2024 inventorship guidance explaining that an AI-assisted invention is patentable so long as a human made a significant contribution to the conception of each claim. Under this framework, an invention could include claim limitations first generated by AI, provided that a human inventor made a meaningful inventive contribution to the claimed subject matter as a whole. This approach relied on the idea that AI could support the inventive process without disqualifying human inventorship.

The USPTO’s 2024 guidance applied the Pannu v. Iolab Corp. factors to evaluate whether a human’s contribution was significant. These factors looked to whether the person contributed in a meaningful way to the conception of the invention and ensured that inventorship was not based on trivial or incidental activity. For many practitioners, this framework provided a workable method for determining inventorship in mixed AI–human scenarios.

In 2025, however, the USPTO withdrew the 2024 framework and issued revised guidance. The Office stated that inventorship must be determined solely under traditional conception principles and did not retain the Pannu-based structure for assessing significant human contribution. Although some commentators view the 2025 guidance as more flexible or patent-friendly, Bitlaw reads it as creating additional uncertainty because it does not explain how to treat claim limitations that were first proposed by an AI system. The revised guidance offers little detail on how to evaluate inventorship when AI meaningfully contributes to the invention, leaving practitioners without clear direction until courts clarify the law.

Artificial Intelligence (AI) Patent Attorney

Please see Dan Tysver's bio and contact information if you need any AI-related legal assistance. Dan is a Minnesota-based attorney providing AI advice on intellectual property and litigation issues to clients across the country.