This article discusses the patentability of thin wrappers on LLMs.

1. Summary

Using existing AI technology with new data was determined to be not patentable in the court case Recentive Analytics, Inc. v. Fox Corp. The Recentive decision adds challenges to patenting AI especially when the invention is a thin wrapper on top of a third party LLM. An area for potential patentability using LLMs is a technological improvement, such as improving how a computer operates or improving a technological process.

2. Recentive Decision: Using Existing AI with New Data is Not Patentable

Recentive Analytics, Inc. sued Fox Corp. for allegedly infringing the following patents:

- U.S. Patent Number 10,911,811

- U.S. Patent Number 10,958,957

- U.S. Patent Number 11,386,367

- U.S. Patent Number 11,537,960

The District Court found that the patents were directed to an abstract idea and thus not patent eligible. The District Court decision essentially means the District Court found the U.S. Patent and Trademark Office (USPTO) should not have issued the patents and dismissed Recentive‘s lawsuit against Fox Corp. In other words, the District Court determined the Recentive patents are worthless and Recentive cannot collect damages from anyone infringing the patents, including collecting damages from Fox Corp.

Recentive appealed the District Court's decision to the Court of Appeals for the Federal Circuit (CFAC). In Recentive Analytics, Inc. v. Fox Corp., No. 2023-2437 (Fed. Cir. Apr. 18, 2025), the CFAC decided:

we hold only that patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under [35 U.S.C.] § 101.

The CFAC designated Recentive Analytics, Inc. v. Fox Corp. as precedential which means the decision applies to all cases. Recentive is case law unless the Supreme Court reverses the CFAC decision or Congress changes the laws to override the decision.

3. Thin LLM Wrappers

A thin wrapper is typically a small amount of software code that relies on a larger existing software package. A thin wrapper on an LLM may be a small amount of software code that relies on the LLM to perform the functionality. As thin LLM wrappers typically apply new data to an existing LLM, patenting thin LLM wrappers can be challenging.

Areas to consider for patentability:

- Does the prompt allow LLMs to perform function(s) that is otherwise not possible without the prompt. For example, can the LLM perform reasoning with the prompt that the LLM otherwise could not

- Do have a new fine-tuning method that improves an aspect of training. For example, faster training time, less memory usage, lower power usage, etc.

- Does the data pipeline improve the LLM from a technology standpoint. For example, does the data pipeline create an environment where there are fewer calls to the LLM that allows an answer to be returned to a user faster.

Although multiple paths to patentability may exist, two potential factors for AI patentability are:

- Improve how computers operate. In Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), the CAFC found Enfish’s self-referential database improved the function of a computer because of "indexing technique that allows for faster searching of data" and "allows for more effective storage of data other than structured text, such as images and unstructured text."

- improved technological process. In McRO, Inc. v. Bandai 837 F.3d 1299, 1315, 120 USPQ2d 1091, 1102 (Fed. Cir. 2016), the CAFC found McRO's patent "improved an existing technological process" and thus is patent eligible. McRO's patent covered using transcribed audio to generate lip movement and facial expression for 3d computer animation. Before McRo's invention, 3d computer animation was performed manually by artists. McRO improves a technological process by allowing a computer to automatically create lip and facial movements that were previously performed manually by artists.

4. Courts Do Not Follow USPTO Guidance

The USPTO has created Patent Eligibility Guidance (PEG) to assist Patent Examiners in determining if an invention is patent eligible or not patent eligible. The 2019 PEG example 39 titled "Method for Training a Neural Network for Facial Detection" includes interactively training a neural network to improve facial recognition in an image. The USPTO describes example 39 as a “neural network is a framework of machine learning algorithms" (emphasis added). One aspect of example 39 is to apply image transformations to the training images to create more training images. Example 39 states a side effect of using the transformed images in training is an increase in false positives with non-facial images. A second aspect of example 39 addresses false positives by "performing an iterative training algorithm, in which the system is retrained with an updated training set containing false positives introduced after face detection was performed on a set of non-facial images" (emphasis added).

Recentive's argument for patentability based "to generat[ing] customized algorithms" did not persuade the CAFC. Regarding Recentive's argument for patentability based on iterative training, the CAFC decision states that "machine learning model [to] be [""]iteratively trained[""] or dynamically adjusted in the Machine Learning Training patents do not represent a technological improvement."

Recentive similar arguments to the USPTO guidance failed to convince the CAFC that the Recentive patents were eligible for a patent. Courts have long taken the position they are not bound by the USPTO guidance. Navigating both the USPTO guidance and the court decisions creates challenges in patenting AI.

4. Future

After the CAFC decision in Recentive Analytics, Inc. v. Fox Corp, using existing AI technology with new data is not patent eligible. In other words, existing AI technology with new data is not patentable.

The patentability of using existing AI technology with new data can change in two situations:

- A Recentive appeal to the Supreme Court where the Supreme Court reverses the CFAC decision in Recentive Analytics, Inc. v. Fox Corp.

- Congress can change the laws to override court decisions. As of the writing of this article, the Patent Eligibility Restoration Act (PERA) bill in Congress is written to remove some of the limits the courts have placed on patentability. One of the limits PERA would remove is the abstract idea doctrine as applied to some software inventions like the Recentive patents.