Are AI and Machine Learning Patents Doomed After Recentive?

On April 18, 2025, the U.S. Court of Appeals for the Federal Circuit affirmed the district court’s dismissal of a patent infringement lawsuit brought by Recentive Analytics against Fox Corporation, holding that the asserted AI and machine learning patents were not patent eligible under 35 U.S.C. § 101. The decision is significant for patent attorneys and applicants in the AI space, particularly those seeking protection for inventions that incorporate machine learning (ML).

Recentive’s four asserted patents involved software for generating event schedules and network maps using machine learning models trained on historical data. Although the applications had successfully overcome  Section 101 rejections during examination before the U.S. Patent and Trademark Office (USPTO), the district court, and now the Federal Circuit, held that the claims were directed to abstract ideas under Alice Step One, and lacked the inventive concept necessary to confer eligibility under Alice Step Two.

In Recentive, the Federal Circuit acknowledged the growing significance of AI and machine learning and emphasized that its holding is limited to generic machine learning applications. However, the broader implications of this decision for existing patents and applications pending before the USPTO remain uncertain.

Simply Training Machine Learning Models is Insufficient to overcome Patent Eligibility

A key fact in the case was Recentive’s own concession: the machine learning models employed were conventional. The Federal Circuit reaffirmed that iteratively training a machine learning model on data does not transform an abstract idea into a patent-eligible invention. Similarly, confining the trained machine learning model to a particular technological field is insufficient unless the implementation introduces a specific, non-generic improvement to computing technology and describes how this improvement is accomplished.

It is important to note that most machine learning models are inherently trained on large, often complex datasets to generate predictions or classifications. However, this alone is routine and well-understood in the field.  Although prior to Recentive one could argue that the trained machine learning model represents a technological improvement, the Recentive decision makes clear that such arguments are insufficient unless the claims specifically describe how the technological improvement is achieved.

Drafting for Eligibility: Lessons from Recentive

The conclusion in the Recentive decision reinforces that AI and machine learning claims remain patent eligible. The focus, however, must shift from the trained machine learning model simply generating a result (i.e., generating schedules or analyzing networks as was the case in Recentive) to how the machine learning model performs the task in a technically novel manner.

To maximize the likelihood that applications and patents will be found eligible under Section 101 by the USPTO and courts, applicants should carefully craft a narrative of a technological advancement during the drafting process. This narrative should be included in the specification and culminate in the claims, clearly demonstrating how the use of AI or machine learning results in a specific technical improvement. A well-developed narrative should:

• Describe a concrete technical problem. The specification should describe a problem that existed in prior systems and processes.
• Present a specific solution. The specific solution can be a novel machine learning model architecture, an inventive feature extraction technique, or an improved training method that avoids bias or overfitting in certain contexts. Simply training the machine learning model is insufficient.
• Highlight how the machine performs differently from a human. The courts have repeatedly held that simply stating that the improvement lies in the machine learning system performing a task faster, efficient, or more accurately than a human is not sufficient. The key is how the machine learning system does it differently. This how should be captured in the claims.
• Demonstrate the result and its technical impact. The result and technical impact of the machine learning system can be reduced system latency, lower energy consumption, or improvements in computing performance. Importantly, these technical impacts should be tied to the invention’s technical features that are captured in the claims.

Pitfalls to Avoid

It is also essential to avoid common pitfalls that undermine patent eligibility. These pitfalls include:

• Using high-level functional language in claims without technical context. Claims that recite steps such as “training a model,” “generating predictions,” or “displaying results,” without grounding these steps in a specific, non-conventional technological framework are unlikely to be found patent eligible individually and as a combination.
• Claiming only the result without technical detail. Outcome oriented claims, such as “updating a network map” or a “schedule” as in Recentive, but without detailing the technical mechanisms behind it, also fail to meet patent eligibility standards. Courts have increasingly held that such outcome-oriented claims are insufficient to confer patent eligibility under Alice Step One. In contrast, as illustrated in McRO, Inc. v. Bandai Namco Games, 837 F.3d 1299 (Fed. Cir. 2016), claims that included specific rules and algorithms for achieving the result were found to be patent eligible because they recited a particular technological implementation rather than an abstract idea.
• Framing the invention around performance improvements. Another pitfall is crafting a narrative that a machine learning model performs a task faster, more efficiently or more accurately than an existing system or a human. While performance gains are valuable in practice, they do not confer patent eligibility. Similarly, the fact that a machine learning model can perform tasks beyond the capacity of the human mind do not necessarily make the invention patent eligible. The courts are clear that such advantages are not technological improvements. Instead, for the claims to be patent eligible, the improvement must lie in how the machine learning system operates differently from what humans or conventional computers do.

Looking Forward: Practical Guidance

Although Recentive offers some guidance for how AI and machine learning inventions may be patent eligible, the legal landscape remains uncertain. The most valuable improvements of AI and machine learning are often tied to training data. However, under Recentive these improvements are unlikely to support the finding of patent eligibility. Given this uncertainty, innovators and patent practitioners should adopt a strategic, multifaceted approach:

• Evaluate all forms of intellectual property protection. If the core innovation lies in training the machine learning system using training data, consider whether trade secret protection is a more appropriate option. Keeping the training data a secret, can offer long-term protection without the public disclosure required by patents.
• Treat USPTO examination as a necessary but insufficient hurdle in overcoming patent eligibility. Success during prosecution does not guarantee success in litigation. Courts need not defer to the USPTO’s determinations on patent eligibility. As Recentive demonstrates, patents that overcome § 101 during examination can still be invalidated in litigation. Applications should be drafted with an eye toward litigation from the outset.
• Use the specification to tell the technical story. Applications should explain the technical context and the deficiencies of existing systems, how the AI or machine learning approach is different and improves upon existing technology. This how should be captured in the claims. Close collaboration with inventors to identify the specific technical improvements is paramount.
• Ensure that claims are not merely descriptive of goals. By reciting specific steps, data structures, or training techniques that provide a clear path from problem to solution, and crafting claims directed at same, goes a long way to preventing patent eligibility issues.

Focus on the How

Recentive serves as a reminder that AI and machine learning, regardless of their sophistication or widespread use, are not automatically patent eligible. The innovation does not lie merely in using or training AI, but in how the technology is specifically implemented to achieve a technical improvement. Applicants would be well advised to treat Recentive as a roadmap for drafting claims and for developing a narrative that supports patent eligibility.

Image Source: Deposit Photos
Author: alexskopje
Image ID: 30864475