Listen to this post

Artificial intelligence (AI) is transforming industries, including the life sciences industry, from drug discovery and diagnostics to medical devices and manufacturing. With that transformation comes a critical challenge: crafting strategies to protect AI-related innovations. Strategies for protecting AI innovations are unique. Approaching IP protections for AI innovations in the same way as approaching IP protections for traditional biotech, diagnostic, or medical innovations is likely to lead to failure and frustration. Understanding the legal framework for AI innovations, how it applies in practice, and what strategies improve the odds of obtaining protection is essential for any company innovating with AI.

This post focuses on patent legal framework and strategies, although copyright, trade secret, trademark, and other forms of protection should also be considered.

To qualify for a patent, an invention must be useful and directed to patent-eligible subject matter under 35 U.S.C. § 101, novel under § 102, and non-obvious under § 103. Eligible subject matter must fall within a statutory category: process, machine, manufacture, or composition of matter, or any improvement to the foregoing. The real obstacle is the judicially created exceptions; an invention directed to a law of nature, natural phenomenon, or abstract idea is ineligible unless the claim as a whole adds significantly more than the exception itself.

Framework for Section 101 Patent Eligibility

The USPTO applies the Alice-Mayo framework in several steps. Step 1 asks whether the claim falls within a statutory category. Step 2A, Prong 1, asks whether the claim is directed to a judicial exception. AI claims typically fall within the abstract idea category of judicial exceptions due to commonly being directed to mathematical concepts, methods of organizing human activity, or mental processes. If the claim is directed to a judicial exception, then the analysis proceeds to Step 2A, Prong 2 to determine whether the claim integrates the exception into a practical application. If the claim does not integrate the exception into a practical application, then Step 2B asks whether the claim includes additional elements amounting to “significantly more” than the abstract idea. There is no bright-line test for “significantly more,” though the USPTO has identified limitations that courts have and have not found sufficient.

Examples of Alice-Mayo Analysis and Outcomes

Real-world examples illustrate the framework in action. U.S. Patent No. 12,268,530 (directed to the Oura® ring) claims a wearable device that measures heart rate variability, feeds data through two machine learning classifiers, and displays an illness-risk metric to the user. This anchors the AI functionality to specific hardware sensors. U.S. Patent No. 12,265,373 (directed to an implant design system) embeds AI in a complete workflow: populating a database with patient anatomy, using an AI model to generate a patient-specific device design, and transmitting it to a 3D printer. By contrast, a claim that merely recites training a neural network and detecting anomalies, as set forth in Example 47 – Anomaly Detection described by the USPTO, without tying those steps to specific hardware, data, or outputs, is far more likely to face eligibility rejections.

Recent USPTO Guidance on AI Patent Eligibility

The USPTO’s 2024 guidance introduced key clarifications aimed at promoting clarity and consistency in the examination process. The USPTO reminded Examiners that they must distinguish between claims that actually recite an abstract idea and claims that merely involve one. Two points matter most for AI applicants: (1) AI limitations that cannot practically be performed in the human mind do not fall within the “mental processes” grouping, narrowing a basis historically used to reject AI claims; and (2) claims reflecting a specific technological solution to a technological problem may qualify as eligible.

These principles were reinforced in Ex Parte Desjardins (2025), where the USPTO Appeal Review Panel vacated a § 101 rejection for a patent on training machine learning models. Although the claims included computing a posterior distribution, which the USPTO determined was a mathematical concept, the panel found them directed to an improvement in machine learning at Alice Step 2. The panel noted that “[c]ategorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology.”

Practical Strategies for Patent Applications Directed to AI Innovations

Application preparation

How you frame an AI invention in a patent application can determine whether it survives examination. Applicants should frame the innovation as solving a specific technical problem with a concrete, real-world result, not merely a business problem. In drafting, applicants should lead with the technical nature of the innovation, tie software claims to specific hardware, data inputs, and outputs, and detail improvements to existing technology in the specification. If eligibility appears difficult, applicants should consider trade secret protection as an alternative.

In the filing process, applicants should also consider art unit targeting. USPTO examiners are grouped into art units with varying allowance rates, and most applications could plausibly fall in more than one. Third-party analytics tools can predict likely assignment, allowing applicants to iterate on the application to obtain a more favorable predicted art unit and potentially avoid § 101 issues at the outset.

Post-filing

After filing, applicants should use examiner statistics (allowance rate, interview success rate, appeal win rate) to understand preferences and likelihood of success. Engaging in interviews early and mapping claims to USPTO examples can be especially valuable. Leverage the framing included in the application as discussed above. Where necessary, applicants can attempt to place claims before a different examiner, though this typically requires multiple applications and a longer time horizon.

Conclusion

Patent protection for AI innovations is achievable but requires deliberate strategy at every stage and traditional approaches to non-AI or non-software patent applications may not be helpful. The USPTO’s 2024 guidance and decisions like Ex Parte Desjardins signal growing recognition that AI innovations deserve protection when they deliver real technical improvements. Companies that frame their inventions carefully, invest in detailed disclosures, and engage proactively with examination will be best positioned to build durable AI patent portfolios.