Introduction
In July 2024 [4], the US Patent and Trademark Office (USPTO) issued guidance to address the patent eligibility challenges associated with AI-related inventions. This guidance aims to clarify the criteria for securing patents in the rapidly evolving field of artificial intelligence, emphasizing the importance of specificity in claims and practical applications [2].
Description
In July 2024 [4], the US Patent and Trademark Office (USPTO) issued guidance aimed at clarifying the patent eligibility challenges associated with AI-related inventions. This guidance includes three written examples with sample claims analyzed for eligibility [4], one of which involves using an artificial neural network (ANN) for data anomaly detection [1]. As AI innovations proliferate across various industries [5], securing patents for these technologies has become increasingly critical [5]. The guidance provides valuable insights into enhancing the likelihood of patent approval by emphasizing the importance of specificity in claims, particularly concerning practical applications and significant technological advancements.
To avoid a Section 101 rejection [2], it is crucial to refrain from including judicial exceptions such as mathematical formulas [2], mental processes [2], or methods of organizing human activity in patent claims [2]. For instance [2], generically claiming stitching without restricting the method to a mathematical calculation can prevent the limitation from being categorized as a mathematical concept [3]. While having a specific hardware implementation [1], such as an application-specific integrated circuit (ASIC) [2], can bolster the patent eligibility of a claim [1], the mere presence of physical structure does not guarantee protection from being categorized as non-statutory [2]. Claims that mix statutory and non-statutory elements may be interpreted as non-statutory under the broadest reasonable interpretation (BRI) [2].
The guidance outlines several key considerations for patent eligibility. Firstly [1], claims that focus solely on the high-level training and application of the trained model are likely to be considered patent ineligible [1]. For example [1], converting a time-frequency representation of a mixed speech signal into embeddings involves a mathematical equation [3], while clustering these embeddings using a k-means algorithm and obtaining masked clusters through binary masks are also considered mathematical calculations [3]. This raises concerns about the novelty and nonobviousness of such claims [3], as they rely on well-known machine learning and data processing methods [3]. Secondly [1], framing claims around the system’s response to the AI’s output may improve patent eligibility [1]. Lastly [1], claims that describe practical applications [2], such as detecting malicious network packets or diagnosing glaucoma risk using AI, must be carefully crafted to avoid detection issues and should not explicitly detail mathematical operations, as those may face rejection.
To enhance the likelihood of patent approval [1] [5], applicants should emphasize the significant technological advancements their AI inventions provide [5], such as introducing new functionalities [5], significantly reducing computing time [5], or improving accuracy [5]. It is essential to articulate the specific mechanisms of the AI [5], rather than simply stating its capabilities [5], and to differentiate it from existing technologies [5]. Tying the AI invention to concrete [5], unconventional real-world applications with defined outputs can strengthen the case for patentability [5]. Specificity in demonstrating how the invention addresses a particular problem is crucial [5].
This update aims to assist USPTO personnel and stakeholders in assessing the subject matter eligibility of claims related to AI technology during patent examination [1], appeals [1] [3], and post-grant proceedings [1]. Additionally, it introduces a new set of examples to aid in the application of the USPTO’s subject matter eligibility guidance for AI inventions [1]. Recent guidance indicates that previous examples [2], such as Example 39 from the 2019 PEG [2], may be less effective in overcoming Section 101 rejections compared to newer examples [2], highlighting the nuanced approach required in drafting claims to navigate the complexities of patent eligibility in the context of AI and related technologies [2]. The guidance’s reliance on hypothetical claims raises concerns about its applicability [3], as these examples may not reflect real-world scenarios and could face challenges under Section 102 and 103 [3]. The lack of detailed descriptions in the hypothetical claims complicates the application of the BRI standard [3], necessitating a comprehensive examination of claims in the context of their entire disclosure [3]. Patent practitioners observe that success rates for AI patents may vary across different art units within the USPTO [5], although comprehensive data on these discrepancies is limited [5]. Innovators are encouraged to adopt an evidence-based approach when describing their inventions to maximize the potential for robust patent protection [5].
Conclusion
The USPTO’s guidance on AI-related patent eligibility has significant implications for innovators and patent practitioners. By emphasizing specificity, practical applications [2] [3] [5], and technological advancements [5], the guidance seeks to streamline the patent approval process for AI inventions. However, the reliance on hypothetical examples and the variability in success rates across different art units highlight the complexities involved. Innovators are encouraged to adopt a strategic [5], evidence-based approach to maximize their chances of securing robust patent protection in this rapidly evolving field.
References
[1] https://www.jdsupra.com/legalnews/uspto-s-ai-patent-eligibility-guidance-4255847/
[2] https://ipwatchdog.com/2025/04/09/ipwensdays-delving-uspto-ai-subject-matter-eligibility-guidance/id=188067/
[3] https://www.llpatent.com/delving-into-the-uspto-ai-subject-matter-eligibility-guidance/
[4] https://ipwatchdog.com/category/ipwensday/
[5] https://www.lexology.com/library/detail.aspx?g=00f1c826-3d0e-4b01-bc58-63ea7cddcbea
