.png)
Startups and early-stage investors often ask the same question: Is our AI or software invention patentable?
The short answer is yes, as long as the invention is novel, non-obvious, and represents a specific solution to a particular technological problem, rather than broadly presenting the idea of a solution.
Abstract ideas and mental processes are excluded from patent protection, and in the past, the USPTO has demonstrated a willingness to use that basis to reject many software inventions, unless careful strategic considerations are taken into account when drafting.
While careful drafting remains a primary concern for AI and software inventions, recent U.S. Patent Office guidance has signaled increasingly favorable views toward AI and software eligibility, and the landscape is shifting in favor of AI and software patents.
This post overviews the general patent examination process, discusses recent USPTO guidance on AI and software inventions, and provides strategic considerations useful for framing software inventions in the most eligible manner.
The Examination Process – Software Considerations
Every patent application is examined for novelty (is it new?) and non-obviousness (is it merely a straightforward combination of known components yielding predictable results?). These baseline requirements apply to all inventions, including physical products and software alike.
Software and AI patent applications face an additional hurdle during examination that physical products typically don't: Subject Matter Eligibility. Abstract ideas, mathematical formulas, and mental processes are excluded from patent protection under the criteria of Subject Matter Eligibility, and many software and AI applications can be characterized as ineligible abstract ideas, mathematical concepts, or mental processes if they are not framed carefully.
To reduce the chances of a Subject Matter Eligibility rejection, the key has traditionally been to frame the software invention as a specific way of solving an identifiable technical problem. Software claims are strongest when they describe inventive subject matter integrated into a practical application, improve the functioning of a computer or system, or address a problem arising from the technology itself. Conversely, software claims are weakest when they broadly describe the idea of solving a business problem with a computer, or broadly describe automating a task that humans have traditionally performed with pen and paper.
The case law around subject matter eligibility is widely recognized as being complex and convoluted, which has made software inventions particularly tricky to pursue. However, recent guidance suggests that the trend may be changing.
Recent USPTO Guidance: The Trend Is in Your Favor
Over the past two years, the USPTO has issued a series of guidance documents and decisions that have meaningfully shifted the landscape in favor of AI and software patents.
The current USPTO Director has been a vocal proponent of "Expansive Eligibility." In testimony before the U.S. Senate, the Director stated that "categorically excluding AI innovations from patent protection in the United States jeopardizes America's leadership in critical emerging technologies,” and later issued guidance to the USPTO warning examiners against overbroad rejections of AI inventions. Separately, in August 2025, the USPTO issued guidance telling examiners not to treat complex AI computations as "mental processes" unless a human could realistically perform them with pen and paper, narrowing the broader manner in which that concept has often been applied.
In September 2025, the Director further issued a rare Appeals Review Panel decision in Ex Parte Desjardins, vacating a Subject Matter Eligibility rejection of claims, instead finding those claims to be Subject Matter Eligible. In that case, the claims were directed to training a machine learning model on multiple tasks while preserving performance on previously learned tasks.
Although the Examiner and Appeals Board found the claims broadly recited abstract ideas and mathematical concepts, the Director reversed the decision noting that the claims represented improvements to how the machine learning model operates. Specifically, the Director noted that the claimed methods protected knowledge about previous tasks to overcome the problem of “catastrophic forgetting” often encountered in continual learning systems, and further reduced storage requirements and complexity for those systems. These improvements to the underlying system were cited to find that the invention was integrated into a practical application and was therefore subject matter eligible.
As a result of these recent actions, the institutional posture towards software and AI inventions is more applicant-friendly than it has been in many years.
What Doesn't Work
Even with the recent favorable guidance, subject matter eligibility remains a key issue for AI and software inventions, so these claims still need to be drafted carefully to avoid eligibility problems.
For instance, in April 2025, the Federal Circuit (the appeals court for patent cases) issued its first major ruling on AI/machine learning patent eligibility. In Recentive Analytics, the court considered claims that were directed to dynamically generating optimized schedules for live events using machine learning models. The court found that the claims merely applied generic machine learning techniques to the fields of event scheduling, and further rejected the eligibility argument that the systems could perform a task previously undertaken by humans with greater speed and efficiency than could previously be achieved. As such, those claims were found to be directed to abstract ideas for which computers were merely invoked as a tool, and thus, were not subject matter eligible. The court emphasized that the claims on appeal did not recite specific improvements to the machine learning models to be applied.
The practical takeaway is that claims should be directed to a specific technological feature or function that enables the AI to be applied to a new field, or a specific technological mechanism for improving the underlying function of the AI or system. Claims that merely describe the broad concept of using AI to accomplish a task are much less likely to prevail.
Why This Matters for Start-ups and Investors
A common pitfall is that founders and technical teams often build genuinely inventive systems, but describe them in broad business or marketing terms that can be read at an abstract level. These teams highlight the usefulness of the outcome rather than providing the technical and engineering details that were required to bring that inventive system into reality.
However, the patent system does not reward describing a desired outcome and its usefulness. In practice, your AI or software invention claim is most patentable where it discloses the precise technical know-how behind how a specific problem was solved, how to enable new functionality that wasn’t previously possible, or how to improve the underlying functioning of the system such as reducing memory requirements or improving computation speed. Getting that framing right at the drafting stage is important because those problems can be difficult or impossible to fix after a rejection.
For diligence purposes, the questions worth asking are practical: Does the patent describe a specific technical implementation, or just a concept? Does the specification explain why this approach is better than what existed before? Are the claims tied to the architecture, or just to the outcome? A patent that answers those questions well is a meaningful asset. One that doesn't is much more likely to face eligibility challenges during examination or enforcement.
Author
Scott Seeley, Registered Patent Attorney
Eastgate IP
eastgateip.com
Resources
