AI Patent Eligibility Has Shifted With One Machine-Learning Case

Over the past few months, the US Patent and Trademark Office has treated artificial intelligence and machine learning inventions more favorably under Section 101 of the US Patent Act than they have been in years.

That shift didn’t happen overnight, but it crystallized through a series of developments centered on the case of Ex parte Desjardins, culminating in new guidance that is reshaping how examiners and applicants alike think about eligibility for AI-related claims.

A Turning Point

The PTO made waves in September 2025 when the Appeals Review Panel, under Director John Squires, vacated a Section 101 rejection in Desjardins. A machine-learning-related application had been rejected on the grounds that its claims were merely abstract.

But Squires agreed with inventor Guillaume Desjardins that the technology would improve “how the machine learning model itself operates, and not, for example, the identified mathematical calculation.”

Appeals Review Panel members agreed, saying the claims recited meaningful technical improvements to how machine-learning models operate, including reducing system complexity, improving multi-task learning, and preventing “catastrophic forgetting.” The panel further explained that such improvements could satisfy the “practical application” requirement under the Alice test for patent eligibility and under Section 101.

From the outset, the message was clear: AI innovations shouldn’t be reflexively categorized as abstract ideas. The panel cautioned that categorically excluding AI innovations from patent protection risks undermining US leadership in critical emerging technologies.

Elevating Desjardins

The PTO took the additional step of designating Ex parte Desjardins as a precedential decision in November 2025. This was more than symbolic—it meant Desjardins was now the binding authority across the PTO and would formally guide both examination and appeal decisions.

It also emphasized a point that practitioners had been arguing for years: If a claim improves the functioning of a computer or another technology, even through logical structures and processes rather than physical hardware, it can be patent-eligible under Section 101.

Such reasoning is grounded in longstanding authority at the US Court of Appeals for the Federal Circuit, such as in the Enfish LLC v. Microsoft Corp. decision recognizing that software-based improvements can be non-abstract.

By elevating Desjardins, the PTO signaled that it wants examiners to assess Section 101 eligibility through the lens of technical contribution rather than reflexive abstraction. Claims that describe specific architectural, performance, or complexity improvements to AI and machine-learning systems are now firmly within the realm of patent eligibility.

A Systematic Change

The USPTO took the next step in December 2025 and issued a formal memorandum integrating Desjardins into the Manual of Patent Examining Procedure, or MPEP. This update is meaningful because it changes how Section 101 is taught and applied in practice.

The MPEP revision instructs examiners to evaluate whether claims recite improvements to computer functionality or other technological fields when assessing eligibility under step 2A, prong two of the Alice framework.

Rather than issuing generic or overly broad abstract-idea rejections, examiners now must look for concrete benefits recited by the claims and supported by the specification. Examples include improved model training efficiency, reduced complexity, better data structures, or other technological advances.

The memo clarifies that eligibility should be aligned with existing guidance. Traditional patentability doctrines such as novelty (Section 102), nonobviousness (Section 103), and written description (Section 112) remain the primary mechanisms for policing claim scope. Eligibility is framed as a technical inquiry into how the invention operates, not as a categorical exclusion of software or AI.

What This Means

Taken together, these three developments mark a significant shift in how the PTO handles subject matter eligibility for AI and machine-learning inventions.

For practitioners and applicants, the lesson is that successful AI patenting is increasingly about claiming and supporting specific technical improvements, not just high-level outcomes or the use of AI in a business context.

Claims that explain how a model is trained, how its architecture improves performance, or how processing is optimized are now more likely to survive Section 101 scrutiny.

This trend is already visible in subsequent decisions by the Patent Trial and Appeal Board. In Ex parte Carmody, the PTAB reversed a Section 101 rejection on Dec. 30 because the claims recited a modular machine-learning architecture with defined training datasets and plug-and-play models that improved how the system operated, not merely what it accomplished.

That reasoning closely tracks Desjardins and underscores the new direction of Section 101 analysis for AI technologies.

Where Things Stand

While Desjardins doesn’t eliminate Section 101 scrutiny entirely, it reframes that scrutiny in a more technology-focused way that is consistent with both Federal Circuit precedent and the PTO’s modernization goals. For innovators in AI, this means that careful claim drafting and robust technical disclosure can shift the eligibility analysis in their favor meaningfully.

In short, Desjardins has evolved from a single appellate intervention into a cornerstone of modern Section 101 practice for AI technologies. Its influence is already shaping PTAB outcomes and will likely continue to do so in 2026 and beyond.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Matthew R. Carey is a partner and chair of the electrical and computer technologies practice group at Marshall, Gerstein & Borun.

Write for Us: Author Guidelines