Recentive: Raising the Patent Eligibility Bar in AI Related Inventions

This post is part of MoFo's 2025 Intersection of AI and Life Sciences blog series. In this blog series, we explore how artificial intelligence is revolutionizing research, innovation, and patient care in the life sciences. Stay tuned for expert insights regarding the impact of AI on intellectual property, licensing, contracts, regulatory policy, enforcement, privacy, and venture markets in life sciences.

AI is being used in the life sciences space to make important discoveries, ranging from new diagnostics to new therapeutic compounds and beyond. In one of the first cases from the Federal Circuit addressing patent eligibility for machine-learning (ML) inventions, the court ruled that applying "generic" ML techniques to a new data environment to automate a task previously performed by humans-in this case "event scheduling"-is an "abstract idea" and thus not patent-eligible, even if the claim includes superficial limitations related to training and improving an ML model. See Recentive Analytics, Inc. v. Fox Corp., No. 2023-2437, slip op. at 18 (Fed. Cir. Apr. 18, 2025).

The Recentive case shows that although inventions relying on ML are patentable, simply using AI to perform tasks that are currently done manually or using generic AI to answer a new question may be unpatentable. It may not be enough to claim a method for sorting through large amounts of sequencing data to find patterns or using a generic LLM to generate a compound that binds to a target of interest.

However, patent protection remains available when applicants disclose and claim an ML invention in sufficient detail, demonstrating that the new ML technique provides a technological advantage. For instance, the Recentive panel emphasized that there remains a path to patent protection for ML-based inventions, noting that:

Machine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology. Today, 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 § 101. Recentive, No. 2023-2437, slip op. at 18.

In light of Recentive, describing the technical advances that modify a generic AI model for a specific task becomes increasingly important. The technological advantage should be beyond increasing efficiency and accuracy over human efforts. Rather, patentability may stand on detailed explanations of how technical steps such as curating the training data or modifying the model's architecture led to the discovery of connections that would not have been possible to find otherwise. For example, technical advances should help the model find connections between gene, compounds, or pathologies that are not apparent on their face.

Overview of Recentive Analytics, Inc. v. Fox Corp.

A representative claim from U.S. Patent No. 11,386,367 that was at issue in Recentive recited iteratively training a machine-learning model to identify relationships between different event parameters and desired outcomes for the events, generating a schedule for a future series of events based on user-defined event parameters and desired outcomes, and updating the generated schedule based on real-time changes to event parameters. While the claim includes several "machine-learning" steps, there is no claim element that informs the reader of any technological improvement beyond conventional ML-techniques; the application of ML in the claims boils down to automation of a process previously performed by a human.

Recentive argued that its claims are patent-eligible because they involve a "unique application of machine learning to generate customized algorithms, based on training the machine learning model, that can then be used to automatically create . . . event schedules that are updated in real-time," asserting that it had "worked out how to make the algorithms function dynamically, so the maps and schedules are automatically customizable and updated with real-time data." Recentive, No. 2023-2437, slip op. at 8, 11-15. However, as conceded by Recentive, iterative training using updated data is "incident to the very nature of machine learning" and "do[es] not . . . mak[e] machine learning better." Recentive, No. 2023-2437, slip op. at 8-12. Because the court viewed the purported advance as nothing more than routine ML behavior applied in a new business context, the court held the claims could not clear § 101 and were thus patent-ineligible.

In sum, Recentive tells patent applicants that:

• Iterative training and real-time updates are not a technical improvement.
• Field-of-use limitations (e., using ML in a new data environment) can't rescue an abstract idea.
• Increased speed and efficiency relative to human-performance of a task are not sufficient for eligibility.

Recentive serves as yet another reminder for patent practitioners and applicants to ensure that their patent applications include robust disclosure of how a claimed invention provides a technical solution to a technical problem. After Recentive, AI inventors still have a clear runway to eligibility: claim either (i) innovations to the ML model itself (e.g., via a novel training process that improves the model for a given task, via a novel data preprocessing technique) or (ii) a concrete technical use of the model's output-and ideally claim both. By anchoring claims in training specifics or downstream control actions, and by disclosing measurable performance gains, applicants can transform generic-sounding AI "ideas" into patent-eligible claims that withstand § 101 scrutiny.

Practitioner & Applicant Takeaways

1. Explain how the model is trained and why those training steps matter. Training details can be critical to patent eligibility. In Recentive, the claims at issue broadly recited "iteratively training" the models "wherein such iterative training improves the accuracy of the ML model." The Federal Circuit balked at this surface-level claiming. Applicants should ensure any training process is described in detail in the specification. In view of Recentive, applicants should also consider including key training steps in the independent claim.
2. Tie the ML model's output to a technical action. A technical action might be one that controls or changes a physical or software process, such as modulating drug-delivery pumps, validating a drug target in vitro, or treating a patient, etc. This is particularly important when the ML model being claimed is a generic or off-the-shelf model. In such cases, the improvement is not to the ML model itself but rather imparted by the downstream practical application of the model's output.
3. Document technical performance improvements. Whenever feasible, ML-directed patent applications should include data showing how the model being claimed does, in fact, provide a technical improvement over the present state of the art. For instance, if the model provides improved image-enhancement abilities, compare the claimed model's performance with performance of state-of-the-art models and include comparison data in the specification. As noted above, automating a manual process previously performed by humans is not enough; there must be a technological improvement, such as an improvement over existing ML techniques, other software, etc.
4. Describe how your ML model/architecture is unique and/or "improved." The Federal Circuit emphasized that Recentive's patents do not claim a specific method for "improving the mathematical algorithm or making machine learning better." Applicants hoping to patent novel models/architectures that improve machine learning should ensure that their applications describe the model/architecture and such improvements in detail.