It is a well-known fact that the Supreme Court’s judgment has been handed down. alice It has become difficult to secure software-related patents. Alice Corporation Pty. Ltd. v. CLS Bank International573 U.S. 208 (2014). Statutory subject matter eligibility criteria are vague and the application of United States Patent and Trademark Office (USPTO) rules is subjective. This increases patent filing costs and frustration, and discourages patent filing. More than 10 years have passed since then. alice In response to the decision, the USPTO issued multiple guidance updates aimed at resolving issues raised by this landmark decision. With each update, the pendulum swings back and forth from inventor-hostile to inventor-friendly when it comes to patent issuance, especially AI-based patent issuance.
For many years, including artificial intelligence in the claims of patent applications seemed to help overcome the problem. alice-Based §101 rejection. under alice Based on the framework, if a claim describes an abstract idea, the next step is to determine whether that abstract idea is integrated into a real-world application. Integration testing can be met in a variety of ways, including “improving computer functionality or other technology or technical areas.” MPEP § 2106.05(a). The USPTO routinely accepts claims that AI improves the capabilities of computers, and has often succeeded in including training of AI models, even though examiners resisted and AI requires training to function.
At some point, the USPTO changed its position and determined that this approach to AI was unsustainable. In July 2024, the USPTO released updated guidance on patent subject matter eligibility specifically focused on AI and numerous eligibility examples illustrating how AI-based applications should be evaluated. This updated guidance and the addition of new examples effectively ended the token use of AI to overcome 101-based denials (i.e., further increased the USPTO’s hostility toward inventors).
Recently, however, the pendulum has begun to swing back toward inventors. First, in August 2025, the new Deputy Commissioner of Patents issued a memorandum that stated:[r]Notes on Evaluating Subject Matter Eligibility of Claims Under 35 USC 101. ” Although this memorandum did not provide new guidelines, it strongly suggests that the USPTO has applied, and continues to apply, its existing framework too harshly against inventors.
Subsequently, a memorandum of understanding was signed in December 2025. EX PART DESJARDINS”, a precedential decision of the Patent Trial and Appeal Board (PTAB) dealing with patent eligibility under §101 of machine learning inventions. The PTAB Panel found that claims directed to methods for training machine learning models are not merely abstract ideas, but provide specific technical improvements, namely overcoming “catastrophic forgetting” and increasing model performance, thus integrating abstract ideas into real-world applications. The December memorandum states the following notice: desjardins It serves as the latest revision of the Patent Examination Procedures Manual, which also states:[t]These updates are not intended to announce new USPTO practices or procedures, but rather to be consistent with existing USPTO guidance. ”
In any case, by acknowledging the eligibility of AI-specific improvements, this shift once again signals a return to a more inventor-friendly treatment of AI-related applications. This interpretation goes beyond previous USPTO examples that focused more narrowly on traditional hardware or operating system improvements. The latest memorandum also expands the scope of what is considered an “improvement in the functionality of a computer,” informing examiners that improving the inner workings of AI systems, such as how machine learning models are trained or how information is retained, should be seen as a practical application and is therefore eligible subject matter.
Although it is still too early to fully understand the practical impact of the December 2025 Memorandum, when combined with the August 2025 Memorandum, it clearly signals the USPTO’s desired transition towards increasing grant rates for AI-based patent applications.
