IP Monday Law Blog 

For companies building real AI and software technology, patent eligibility under §101 has been a persistent frustration. Innovations grounded in model training, data handling, or system architecture are often dismissed as “abstract” simply because they involve algorithms or math.

A recent precedential decision from the United States Patent and Trademark Office's (USPTO) Appeals Review Panel signals a meaningful adjustment in how those innovations are evaluated. The case—Ex parte Desjardins—does not change the law, but it clarifies how examiners must apply it, particularly for AI and machine-learning technologies.

What Changed (What Didn’t)

The USPTO reaffirmed a critical principle: Claims are not ineligible merely because they involve mathematical operations. What matters is whether the claimed system improves how the technology itself operates.

In Desjardins, the claims survived eligibility because they described and claimed technical improvements to the machine-learning system itself—specifically:

• Reduced storage requirements,
• Reduced system complexity, and
• Preservation of prior learning when training on new tasks (a known technical problem in AI systems).

Those are not business results; they are engineering outcomes.

Why This Matters for Startups and Growing Companies

Big companies can absorb repeated rejections. Startups and small & medium-sized businesses (SMBs) cannot. The lesson from Desjardins is practical and actionable:

1. “Better performance” is not enough. Eligibility turns on how your system achieves its result, what tradeoff it makes inside the machine.
2. Your claims must reflect the improvement. A strong specification helps, but the claims themselves must force the system to behave differently.
3. Math is acceptable when it controls the system. Algorithms are not the problem, untethered algorithms are.
4. Examiners must do more than label components “generic.” High-level dismissals without analysis are no longer sufficient.

One Important Warning

The claims in Desjardins still failed under §103. The USPTO was explicit: §101 should not be used as a shortcut. Obviousness and disclosure remain the real gatekeepers.

Bottom Line

For AI and software companies, the message is clear: Describe—and claim—the engineering compromise your system makes. That compromise is often the patentable advance.

If you’re building AI products and relying on IP as part of your growth or investment story, this is a decision worth understanding and using. If have further questions or concerns about how this development may affect your organization, contact IP attorney Mike Murshak.