IP strategies for AI-minded life sciences companies

Life sciences companies are increasingly using AI in wide-ranging ways, including protein design, clinical trial analysis, and disease diagnosis. Although AI-related patent filings in the USPTO have accelerated, it is increasingly difficult to get an application granted. See 2024 AI Index Report, Chapter 1.2 (https://aiindex.stanford.edu/wp-content/uploads/2024/04/HAI_2024_AI-Index-Report.pdf). In view of these trends, we explore considerations for life sciences companies to factor into their patent strategies, including which AI-related inventions should be patented, potential challenges to patenting, and whether trade secret protection may be appropriate.

INVENTORSHIP

Only natural persons can be inventors on a patent. Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022) (cert. denied). Interim USPTO guidance from February 2024 explains that a human using AI can be an inventor if they "significantly contributed to the claimed invention," the test historically applied to determine joint inventorship. 89 FR 10043; Pannu v. Iolab, 155 F.3d 1344, 1351 (Fed. Cir. 1998). "[M]erely recognizing a problem or having a general goal or research plan [] does not rise to the level of conception" and "a natural person who merely recognizes and appreciates the output of an AI system" is not necessarily an inventor. According to USPTO Example 2: Developing a Therapeutic Compound for Treating Cancer, building a machine learning model based on previous experiments or applying an off-the-shelf model to a problem may not rise to inventorship, whereas structurally modifying compounds predicted by an off-the-shelf model may. To avoid an inventorship problem, companies should therefore ensure that an invention arises from more than simply employing AI tools. Companies should also clarify inventorship or ownership relationships with third-party model developers directly in governing contracts.

SECTION 101

Subject matter eligibility is a significant hurdle for AI patents, which are often heavily dependent on mathematical algorithms--considered abstract ideas. The life sciences industry faces additional §101 hurdles, particularly for inventions that diagnose a natural phenomenon using conventional means of detection. See Athena Diagnostics v. Mayo Collaborative Servs., 915 F.3d 743 (Fed. Cir. 2019).

The case law is developing, but AI-related patents are being rejected under §101. In Realtime Data v. Array Networks, 2023 WL 4924814 (Fed. Cir. 2023), the majority found that claims relating to methods of selectively compressing files were directed to the abstract idea of "data manipulation" and were ineligible because they did not specify "any particular technique" for data compression or "the particular rules" for producing a smaller set of data from a larger one. By contrast, claims directed to an improvement to otherwise-known laboratory processes through specific application of mathematical algorithms have been found patent-eligible. XY v. Trans Ova Genetics, 968 F.3d 1323, 1331 (Fed. Cir. 2020). Claim eligibility often turns on whether the claims describe an improvement to a technology or laboratory technique, and claims should be drafted with that in mind.

SECTION 112

Written description and enablement are also important considerations for AI-related inventions given the difficulty of describing how AI models and technologies work, such that a person of ordinary skill (POSA) could practice the invention or understand the inventor possessed it. In the life sciences context, inventors may lack software expertise. Further, the specification may need to disclose otherwise-confidential algorithms and training data to enable others to practice the invention.

Courts are beginning to consider AI-related inventions through a §112 lens. Judge Newman's dissent in Realtime Data postulated that enablement is "better suited" than §101 to determine the validity of AI-related claims, consistent with a results-oriented claim being analyzable in view of whether the specification appropriately explains how to achieve that result. In Impact Engine v. Google, 624 F. Supp. 3d 1190, 1195 (S.D. Cal. 2022) (appeal pending), the Court found the specification's description insufficient because it failed to explain how the claimed "compiler" performed operations unknown to a POSA. AI-related claims, especially those incorporating functional results, should specify how to achieve the results, and the specification should be detailed in areas that are not known to a POSA. For AI-related inventions in the life sciences, a POSA could be found to require both software and life sciences experience.

ARE TRADE SECRETS A VIABLE ALTERNATIVE?

An AI invention that cannot be described at the level necessary for §101 and §112 may still be protected as a trade secret. The Defend Trade Secrets Act (DTSA) broadly covers "all forms and types of financial, business, scientific, technical, economic, or engineering information" including formulas and programs. 18 U.S.C. §1839(3). In Apex.AI v. Langmead, 2023 WL 3391962, at *3 (N.D. Cal. May 10, 2023), the Court found that Apex.AI's software tools and source code for autonomous vehicles were protectable trade secrets. However, an AI-assisted invention must not be readily ascertainable by another and must be protected by reasonable measures to be maintained as a trade secret. §1839(3)(A). If a competitor legally ascertains the trade secret through reverse engineering or independent development, the trade secret cannot be enforced.

Further on enforcement, many jurisdictions require a trade secret to be identified with sufficient particularity to put the defendant on notice of the claim. See, e.g., Alta Devices v. LG Electronics, 343 F. Supp. 3d 868, 881 (N.D. Cal. 2018). Identification does not mean describing the trade secret to the extent required by §112, but if the trade secret is not capable of being identified and used by the misappropriator, proving damages may be difficult. Further, unlike patent infringement, misappropriation requires acquisition through improper means, which poses additional hurdles.

Though the DTSA does not analyze inventorship, the company must be considered "an owner" of the trade secret. §1836 (§1839(4) defines "owner" as "the person or entity in whom or in which rightful legal or equitable title to, or license in, the trade secret is reposed"). Thus, when employing third-party AI tools, companies should secure ownership or, at minimum, license know-how arising from the application of AI tools and resulting outputs, and negotiate rights beforehand.

In sum, life sciences companies using AI tools should review their patenting practices through the lens of challenges relating to protection, validity, and ownership of patents and trade secrets.

Irene Yang is a partner, and Brooke Böll is senior managing associate at Sidley Austin LLP. Chelsea Himes , a law clerk at Sidley Austin LLP, contributed to the article.