AI-Assisted Drug Discovery After USPTO Inventorship Guidance Updates: Patentability, Conception, and Trade Secret Strategy

Introduction 

The rapid emergence of AI/ML tools in drug discovery, from target identification to generative molecular design, lead optimization, and repurposing, has transformed how life sciences companies build some of their most valuable assets. As AI/ML tools take on an increasingly significant role in the drug discovery process, companies face a set of largely unresolved IP challenges that require careful planning and proactive governance.  

1.Patentability of Discoveries 

The central tension of AI/ML in drug discovery is straightforward: AI/ML is doing more of the work, and in some instances more of the creative work, than ever before, but the law still requires a human inventor to establish patentability. The Patent Act establishes that a patentable invention be directed to eligible subject matter, and that it be useful, novel, and non-obvious. See 35 U.S.C. §§ 101–103. Critically, the invention must be conceived by a human being. See 35 U.S.C. §§ 100(f), 115. The U.S. Court of Appeals for the Federal Circuit confirmed this requirement in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), cert. denied, 143 S. Ct. 1783 (2023), holding that AI cannot be a named inventor under the Patent Act. The question of how much a researcher can rely on an AI/ML tool and still qualify as an inventor, however, is nuanced and fact-dependent. 

2. Inventorship After Thaler v. Vidal 

Following Thaler, the U.S. Patent and Trademark Office (USPTO) addressed AI and patent inventorship on several occasions. In February 2024, the USPTO issued guidance applying the so-called Pannu factors, developed in the context of joint inventorship, to assess whether a human’s contribution to an AI-assisted invention was “significant.” See Pannu v. Iolab Corp., 155 F.3d 1344, 1351 (Fed. Cir. 1998). On November 28, 2025, the USPTO rescinded that guidance in its entirety, pursuant to Executive Order No. 14179, 90 Fed. Reg. 8741 (Jan. 31, 2025). Under the current framework, the USPTO treats AI/ML systems as tools, analogous to laboratory equipment, and it imposes no distinct analytical framework for evaluating AI-assisted inventions beyond existing patent law requirements. 

3. Return to First Principles: Conception by a Human 

What did not change after the Thaler decision or the USPTO’s rescission of its prior guidance on AI-assisted inventions is the requirement that the inventive concept originate with a human. Under well-settled patent law, conception is defined as “the formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention.” Burroughs Wellcome Co. v. Barr Labs., Inc., 40 F.3d 1223, 1228 (Fed. Cir. 1994) (internal quotation marks omitted). The standard requires “a specific, settled idea, a particular solution ... not just a general goal or research plan.” Id. 

While AI/ML can assist a human inventor, the conception of the invention must occur in the mind of a natural person, and, in those instances, only a human can be named as the inventor on a patent application.  

For life sciences companies deploying AI/ML in drug discovery, the most consequential legal question is not whether AI can be a named inventor — it cannot — but whether a human researcher working alongside an AI/ML tool has conceived of the resulting invention. 

Applying this standard to AI-assisted workflows can be challenging, and the answer is highly fact-specific. At the lower-risk end, AI may function as a search or filtering tool, helping researchers navigate large datasets or existing concepts. In this scenario, the human researcher directs the inquiry, evaluates the results, and makes substantive creative decisions. At the higher-risk end, a generative AI/ML model may produce candidate molecules, propose synthetic pathways, or identify biological targets — outputs that may, in some instances, constitute the inventive contribution itself. Between these possibilities lie a wide range of hybrid workflows, such as where a researcher conceives of a target or hypothesis, but the AI/ML model’s output refines or redirects that idea before it is reduced to practice. 

If no natural person can be identified as having conceived of the claimed invention, the resulting patent faces existential vulnerabilities: the claims are subject to rejection, invalidity, or unenforceability for improper inventorship under 35 U.S.C. §§ 101 and 115.  

Mitigating Lack of Human Conception Risk: To reduce the risk that an invention may be deemed “conceived” by AI/ML, companies should require researchers to contemporaneously document their contributions at each stage of an AI-assisted workflow, including the hypothesis or design constraint formed prior to execution of the model, the prompt engineering inputs, and the outputs generated. Companies should also document the selection, modification, and validation decisions made based on AI-generated outputs. This documentation should name the individuals responsible for each substantive decision and should be treated with the same discipline as laboratory notebooks: dated, version-controlled, and tied to specific discovery events. A clear, stage-by-stage record of human direction and judgment provides the evidentiary foundation needed to defend inventorship if a patent is later challenged. 

4. Impact on Patentability 

Another emerging IP question is whether AI/ML-assisted drug discovery creates new hurdles to patentability. AI/ML tools can screen millions of compounds in just days, compressing timelines in ways that may make resulting discoveries easier to challenge as obvious under 35 U.S.C. § 103, while also enabling the deposit of vast volumes of machine-generated compound data into publicly accessible databases that may constitute anticipatory prior art under 35 U.S.C. § 102. To date, neither of these issues has been definitively addressed by USPTO or the federal courts, but each warrants close attention as AI’s role in drug discovery expands. 

5. Trade Secret Protection as a Complementary Strategy 

In addition to patents, life sciences companies often rely on trade secrets to protect their innovations. Trade secrets have no fixed term, require no public disclosure (secrecy is essential), and can protect valuable information, including manufacturing processes, discovery platforms, assay and formulation techniques, proprietary datasets, and other forms of know-how. Such information may not be patentable, for example, due to the inventorship issues discussed above, or owners may otherwise choose not to disclose it. However, the use of AI/ML tools in research workflows creates distinct vulnerabilities that companies must address proactively. 

The Defend Trade Secrets Act (DTSA), the primary federal statute governing trade secret misappropriation in the United States — and its state law counterparts, most of which are modeled on the Uniform Trade Secrets Act — require that a trade secret owner take “reasonable measures to keep such information secret.” 18 U.S.C. § 1839(3)(A). In AI-assisted drug discovery, a wide range of proprietary information may qualify as a trade secret candidate, including model inputs, prompts, training data, fine-tuned model weights, and structure-activity relationships. If such information is transmitted to a third-party AI vendor whose platform uses customer inputs to train or improve its models, a challenger may argue that the owner failed to maintain reasonable measures. Such a failure could potentially defeat protection under the DTSA. A company must therefore ensure that AI/ML tools used by researchers operate within closed, enterprise-grade environments (not publicly available or consumer-facing platforms) where data confidentiality is maintained and inputs cannot be used to train the models. 

Mitigating Risks of Losing Trade Secret Protections: To mitigate these risks, companies should ensure that a data processing agreement or equivalent contractual arrangement is in place with each AI vendor before any proprietary information is shared. Key terms should include restrictions on vendor use of customer data for training or model improvement (opt-in only), zero-retention processing obligations, confidentiality and nondisclosure provisions, security controls, audit rights, and breach notification requirements. These safeguards should be tailored to the sensitivity of the information and reviewed regularly as vendor practices evolve. 

Companies should also integrate these risks into their broader AI governance frameworks. Internal policies should designate which AI/ML tools are approved for use with proprietary data and expressly prohibit the use of publicly available or consumer-facing AI platforms for confidential research. These internal controls complement the contractual safeguards described above and strengthen the company’s position that it has maintained "reasonable measures" to preserve secrecy under the DTSA or applicable state law. 

Conclusion 

AI/ML tools are transforming drug discovery, but the legal frameworks governing patents and trade secrets have not changed to accommodate them. Inventorship still requires human conception. Trade secret protection still requires reasonable measures to maintain secrecy. Companies that fail to address these requirements proactively risk losing protection for their most valuable assets, whether through an inventorship challenge that renders a patent invalid or a trade secret claim undermined by insufficient safeguards over how proprietary information is used. 

Life sciences companies investing in AI-assisted drug discovery should invest equally in the governance, documentation, and contractual infrastructure needed to protect the resulting IP. The measures outlined here can serve as the evidentiary foundation on which patent validity and trade secret protection will depend.