Protecting Innovation in an AI-Powered Age: Patents

With the advent of generative artificial intelligence (or “GenAI”), many companies have begun to shift their innovation strategies to incorporate and rely on GenAI tools. These tools can be powerful drivers of technological innovation, but their use can affect whether the resulting innovations are protectable. Companies using GenAI tools should carefully consider the implications of relying on GenAI technology and implement guidelines as part of their innovation process to document, as appropriate, (1) which AI tools are being used (e.g., tools from decades past like autocomplete or spam filtering or newer “generative” tools); (2) how these tools are being implemented; and (3) what is done with the output of these tools. When developed with appropriate precautions, patents can remain an effective legal tool for protecting AI innovations.

Here, we explore key legal considerations and offer best practices for companies seeking to patent innovations related to or generated in part by AI.

Key legal considerations

As with any emergent technology, AI — and specifically GenAI —has the potential to affect a wide swath of patent law precedent. Here, we identify some key patent law principles implicated by AI that must be considered when crafting policies to protect innovation, specifically inventorship and patent eligibility. Inventorship issues often arise where an inventor attempts to patent a technology that was developed with the assistance of AI, whether or not the technology itself is AI-related. Patent eligibility issues often arise where an inventor attempts to patent AI technology itself.

Inventorship

AI systems cannot be named inventors

The U.S. Court of Appeals for the Federal Circuit — the highest court to have addressed inventorship in the AI context to date — has held that inventorship is limited to natural persons. In Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022) , cert. denied, 143 S. Ct. 1783 (2023), the applicant had listed DABUS, an AI system, as the sole inventor on a patent application. Id. at 1209-10. The United States Patent and Trademark Office (USPTO) rejected the application for lack of valid inventorship and the Federal Circuit agreed. Id. The Federal Circuit found that the Patent Act requires inventors to be natural persons (i.e., human beings). Id. at 1210.

The court held that the term “individual” in the Patent Act referred to human beings unless there is some indication that Congress intended a different reading, and nothing in the statute provided such an indication. Id. at 1211 (analyzing various sections of Title 35); see 35 U.S.C. § 100(f) (defining “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention”); Id. § 100(g) (defining “joint inventor” as “any [one] of the individuals who invented or discovered the subject matter of a joint invention”).

Indeed, the court held that § 115(b)(2)’s use of the personal pronouns “himself” and “herself” rather than “itself” was evidence that Congress intended only human inventors and did not intend for non-human inventors. Thaler, 43 F.4th at 1211; 35 U.S.C. § 115(b)(2) (requiring that an inventor submit a statement when applying for a patent indicating that “such individual believes himself or herself to be the original inventor or an original joint inventor of a claimed invention in the application”) (emphasis added). The court therefore concluded that “Congress has determined that only a natural person can be an inventor, so AI cannot be.” Thaler, 43 F.4th at 1213.

AI systems can be used by human named inventors

In February 2024, the USPTO issued inventorship guidance for AI-assisted inventions that reiterated the central holding of Thaler — i.e. only natural persons can be inventors. Inventorship Guidance for AI-Assisted Inventions, 89 Fed. Reg. 10043 (Feb. 13, 2024) (“Inventorship Guidance”). In addition, however, the Inventorship Guidance addressed a gap in Thaler, confirming that patent protection can be pursued for inventions made by human beings with the assistance of AI . Id. at 10045.

The USPTO clarified that AI-assisted inventions are not categorically unpatentable. Id. at 10046. “[T]he USPTO recognizes that while an AI system may not be named an inventor or joint inventor in a patent or patent application, an AI system — like other tools — may perform acts that, if performed by a human, could constitute inventorship under our laws.” Id. at 10045. Accordingly, the Inventorship Guidance establishes that patent protection may be sought for inventions in which a natural person provided a “significant contribution” to the invention.

The Inventorship Guidance directs patent examiners to apply the factors enumerated in Pannu v. Iolab Corp., 155 F.3d 1344, 1351 (Fed. Cir. 1998), to determine whether a human inventor contributed significantly. These factors investigate whether a named inventor:

1. Contributed in some significant manner to the conception or reduction to practice of the invention;
2. Made a contribution to the claimed invention that is not insignificant in quality when that contribution is measured against the dimension of the full invention; and
3. Did more than merely explain to the real inventors well-known concepts and/or the current state of the art.

Inventorship Guidance, at 10047 (citing Pannu). While listing the Pannu factors, the Inventorship Guidance acknowledges that “a significant contribution to reduction to practice of an invention conceived by another is not enough to constitute inventorship,” noting that a reference in the Inventorship Guidance to a known doctrine “does not imply that reduction to practice is sufficient for invention or is a substitute for conception.” Id.

Recognizing the difficulty of determining inventorship where an AI system was used during the inventive process, the Inventorship Guidance further provides the following non-exhaustive list of guiding principles that can help inform the application of the Pannu factors in the AI context:

• The use of an AI system does not negate a natural person’s contribution as an inventor; a natural person can be listed as an inventor only if they make a significant contribution to the AI-assisted invention.
• Feeding a general prompt for a general problem into an AI system is not a significant contribution. However, a significant contribution could be shown by the way a natural person constructs the prompt in view of a specific problem to elicit a particular solution from the AI system.
• A natural person who recognizes and appreciates the output of an AI system as an invention is not necessarily an inventor. However, a natural person who takes the output of an AI system and makes a significant contribution to it may be a proper inventor.
• A natural person who designs, builds, or trains an AI system in view of a specific problem to elicit a particular solution could be an inventor where the designing, building, or training of the AI system is a significant contribution to the invention created by the AI system.
• Owning or maintaining “intellectual domination” over an AI system alone does not make a natural person an inventor of any inventions created by the AI system.

Id. at 10048-10049.

Patent eligibility

Since Alice Corp. v. CLS Bank, Int’l, 573 U.S. 208 (2014), courts have scrutinized computer- and software-based patents for patent eligibility, requiring patents to meet the two-prong Alice test: A patent is deemed patent ineligible under § 101 of the Patent Act if (1) it is directed to an abstract idea; and (2) it does not contain sufficient additional elements — i.e., an inventive concept. Id. at 217-18. Given the computer and software focus of AI-related innovations, they have become frequent targets of § 101 challenges alleging that they are directed to abstract ideas without inventive concepts.

Courts have addressed these challenges in a variety of circumstances:

Bare application of AI

Courts have found that the use of AI in an invention without further explanation of the particular AI features or algorithms is directed to an abstract idea. In IBM v. Zillow Grp., Inc., 2022 U.S. Dist. LEXIS 41831 (W.D. Wash. Mar. 9, 2022), the claim at issue was directed to a graphical user interface that performed searches and selections. The court found that the claim was directed to an abstract idea because “[t]he claim language is entirely result-oriented, specifying what data enters and leaves the proverbial "black box," but revealing nothing about the inner workings of the box itself.” Id. at *31-32.

Methods of organizing human behavior

AI claims in business-to-business applications may be found to be directed to abstract ideas if they merely optimize economic transactions. In Quad City Patent, LLC v. Zoosk, Inc., 498 F. Supp. 3d 1178 (N.D. Cal. 2020), the claim at issue related to an AI system that used data mining to optimize business decisions by performing statistical analysis. The court found that the claim was “directed to the abstract idea of matching service offers and requests using standardized terms. Such basic economic practices are routinely found to be patent ineligible.” Id. at 1185.

Mental processes

AI claims often fail the patent eligibility test when the AI merely emulates how humans think and solve problems, even if the AI is faster and more efficient than a human. The court in IBM v. Zillow Grp., Inc., 2022 U.S. Dist. LEXIS 41831, at *31-32 (W.D. Wash. Mar. 9, 2022), explained that the claim at issue there was directed to an abstract idea because “[the processes] can be performed with a pen and paper, albeit not with the speed of a computer, and they are focused on the intangible of information.”

Mathematical relationships and formulas

Courts may find AI claims directed to abstract ideas when the claim language implies that the machine learning processes are simply mathematical relationships and formulas. In Purepredictive, Inc. v. H20.AI, Inc., 2017 U.S. Dist. LEXIS 139056 (N.D. Cal. Aug. 29, 2017), the claim at issue related to a machine learning process for providing insights into a business’ data through the use of predictive modeling. The court found that “[t]he method takes the learned functions, evaluates their effectiveness, and selects those most effective to create a rule set. These are mathematical processes that not only could be performed by humans but also go to the general abstract concept of predictive analytics rather than any specific application.” Id. at *15.

USPTO patent eligibility guidance

In July 2024, the USPTO issued updated patent eligibility guidance to aid patent examiners and applicants in assessing the patent eligibility of AI inventions. 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, 80 Fed. Reg. 58128 (Jul. 17, 2024) (“Patent Eligibility Guidance”). The Patent Eligibility Guidance focuses on Step 1 of the Alice framework and includes several examples of claims to AI inventions that are and are not “directed to” abstract ideas.

First, the Patent Eligibility Guidance makes a distinction between claims that “recite” an abstract idea and those that merely “involve” or are “based on” an abstract idea, with claims that recite abstract ideas being more likely to be found to be “directed to” abstract ideas under the Alice framework. Id. at 58134. USPTO patent examiners determine whether a claim recites an abstract idea by (1) identifying the specific limitation(s) in the claim under examination that the examiner believes recites an abstract idea, and (2) determining whether the identified limitations(s) fall within at least one of the groupings of abstract ideas” (i.e., mental processes, mathematical concepts, or methods of organizing human behavior). Id. at 58135.

The Patent Eligibility Guidance provides the following example of an AI claim that does not recite an abstract idea:

• A claim to an application-specific integrated circuit (ASIC) for an artificial neural network, the ASIC comprising: a plurality of neurons organized in an array, wherein each neuron comprises a register, a processing element and at least one input, and a plurality of synaptic circuits, each synaptic circuit including a memory for storing a synaptic weight, wherein each neuron is connected to at least one other neuron via one of the plurality of synaptic circuits. (Claim 1, Example 47 of Patent Eligibility Guidance.)

Second, the Patent Eligibility Guidance clarifies that a claim that integrates a judicial exception (e.g., abstract idea) into a practical application is not “directed to” a judicial exception even if it recites a judicial exception. Id. at 58136. USPTO patent examiners evaluate integration into a practical application by (1) identifying whether there are any additional elements recited in the claim beyond the judicial exception(s), and (2) evaluating those additional elements individually and in combination to determine whether they integrate the exception into a practical application of that exception. Id. One way applicants can demonstrate integration of a judicial exception into a practical application is to show that the claimed invention improves the functioning of a computer or other technology or technological field. Id.

The Patent Eligibility Guidance provides the following examples of claims that improve technology and are not directed to judicial exceptions:

• A claim to a method of using an artificial neural network to detect malicious network packets that recites the additional elements of “(d) detecting a source address associated with the one or more malicious network packets,” “(e) dropping the one or more malicious network packets,” and “(f) blocking future traffic from the source address.” (Claim 3, Example 47 of Patent Eligibility Guidance.)
• A claim to an AI-based speech separation method that recites the additional elements “(f) synthesizing speech waveforms from the masked clusters, wherein each speech waveform corresponds to a different source s_n,” and “(g) combining the speech waveforms to generate a mixed speech signal x' by stitching together the speech waveforms corresponding to the different sources s_n, excluding the speech waveform from a target source ss such that the mixed speech signal x' includes speech waveforms from the different sources s_n and excludes the speech waveform from the target source s_s.” (Claim 2, Example 48 of Patent Eligibility Guidance.)

Best practices for obtaining AI patents

Companies should be deliberate about what AI functionality they use as part of their innovation strategy, how they use that AI functionality, and how they document that use. Likewise, companies should follow the recent jurisprudence on patentability of AI-focused inventions, making sure to draft claim language and identify technological advancements with an eye toward protecting their intellectual property (IP). Whether each of the recommendations below is appropriate varies based on industry and individual circumstances, but internal legal teams should carefully scrutinize their processes to determine whether each is applicable.

Recording the use of AI through invention disclosure forms and interviews

In general, and where applicable, companies prosecuting patents should devote at least some resources to tracking the development of technology and innovation. This can be useful with respect to IP development (e.g., to support inventorship and provide written description support), as well as for general business functions (e.g., to track the use of resources).

Tracking technology and innovation development can take many forms. These can be daily requirements (e.g., requiring daily note keeping) or they can be milestone-based (e.g., requiring inventors to complete a written invention disclosure form (IDF) or verbal interview when they complete an invention). Where AI is used significantly in the innovation development process, companies should carefully consider where and how to document AI-related functionality.

Invention disclosure forms. IDFs may include some of the following questions:

• During which phases/milestones of the project was AI used?
• For each (significant) use of AI:
  • Which AI tools were used?
  • What was the content of the prompts or human input?
  • Who developed the prompts or human input?
  • What were the outputs?
  • How were the outputs used?
  • Who used the outputs?
  • How were the outputs modified, adapted, altered, or otherwise implemented as part of the development process?

Inventor interviews. Inventor interviews may include the above questions on whether AI was used in the development of the invention. When interviewing inventors, interviewers should review the AI-related answers the inventor provided in the IDF (if any) and use the interview as an opportunity to examine the specifics of the AI tools used, how they were used, and how the human inventor contributed. Critical analysis should be focused on usage of AI that affects the more substantial portions of the invention. For example:

• How were the AI tools that were used selected, and who made that selection? Why?
• How were the inputs selected? Were there special considerations from the inventors that led to the selection of the inputs?
• How were the results of the AI tools implemented? Did the inventors need to sort, exclude, or alter results from the AI?
  Information about the inventors’ human contributions to the invention can help support patentability at the prosecution stage.

Duty of disclosure. Tracking and documenting the use of AI in the R&D process might also be important in light of the duties of candor and good faith in dealing with the USPTO. 37 CFR 1.56. These duties are commonly referred to as the “duty of disclosure” and require applicants to disclose all information known to them that is material to patentability. Information is material to patentability when it:

1. Establishes, by itself or in combination with other information, a prima facie case of unpatentability, or
2. Refutes, or is inconsistent with, a position an applicant takes in (a) opposing an argument of patentability relied upon by the USPTO, or (b) asserting an argument of patentability

In applications for AI-assisted inventions, information that raises a prima facie case of unpatentability could include, for example, evidence that demonstrates that a named inventor did not significantly contribute to the invention. Inventorship Guidance for AI-Assisted Inventions, 89 Fed. Reg. 10049 (Feb. 13, 2024). Applicants should thus conduct a reasonable inquiry into whether and how AI was used in the invention process, including assessing whether the contributions made by natural persons are significant enough to establish inventorship, and disclose the results to the USPTO if necessary.

At the time of publication, the authors are not aware of any U.S. District Court case considering the duty of disclosure with respect to GenAI.

Identifying subject matter more likely to be deemed patentable

While the USPTO has announced that there is no per se bar to patentability of AI-related inventions, recent patent eligibility challenges demonstrate that certain subject matter might be more likely to be deemed eligible.

Technological solutions to technological problems. When determining patent eligibility, patent examiners at the USPTO consider whether the claims purport to improve the functioning of the computer itself. This consideration is often referred to as “the search for a technological solution to a technological problem.” MPEP § 2106.05(a). Applicants can thus improve their chances of patenting AI innovations by focusing on their technical advantages. For example:

• Does the invention use less computing power?
• Does it generate outputs with reduced latency or increased efficiency?
• Is it particularly suited for a particular implementation (e.g., mobile, server)?
• Does it claim a specific, hardware-based structure?

To reduce the risk of § 101 rejections, applicants should ensure that they articulate the improvement over the prior art and link that improvement to their patent claims as applicable.

Specific applications of AI to the technology. Similarly, patents that are directed to applications of AI as implemented are more likely to be found patent eligible. For example, the USPTO has granted patents related to:

• Hardware supporting GenAI (e.g., GPUs or TPUs) that is configured to support it in specific mobile, server, or other applications;
• Devices implementing GenAI tools that are specifically designed for that use (e.g., headphones that translate languages in real time); and
• Specific AI software implementations (e.g., AI model structures, including specific descriptions of the system organization and implementation).

Internal education for technical teams

While appropriate procedures are critical for maintaining protectable IP, internal education may be necessary to ensure that employees are being mindful of their use of GenAI. Companies seeking to patent their AI innovations can set themselves up for success by educating their internal technical teams on a few AI best practices. For example, internal education could cover the basics of inventorship and patent eligibility, the importance of tracking which tools the teams use (e.g., third party tools vs. in-house tools), procedures for recording prompts and outputs, and how to document current and planned uses of AI in the development process. Such information can be included in onboarding and training materials on a need-to-know basis and can be tailored to the needs of the team.

GenAI has ushered in a sea change in the way companies develop technology, address problems, and innovate. While this change can be powerful for economic growth, it has significant effects on patent protections. Companies should make sure to take appropriate measures to secure ownership of any new technological developments and continue to innovate with an eye toward protecting their innovation through IP.