Recent developments in AI-assisted research highlight the importance of documenting human involvement to secure IP rights.
Anthropic recently announced that it used a team of Claude agents to discover a reverse transcriptase enzyme system associated with CRISPR-like DNA repeat sequences.[1] While the system hasn’t yet been fully characterised, Anthropic believes it could be used as a new gene editing tool.
This is the first finding to come out of Anthropic’s recently announced AI-guided wet lab,[2] and the company has said that human involvement was “limited to the initial prompt and the lab work”.[3] Pharma companies are also racing towards AI-automated science (that is, where the model proposes a hypothesis; performs the research or employs robotic systems to do the wet-lab experiments; and uses the resulting data for iterative hypothesis improvement). Last week, Roche announced that it was building an AI-automated lab with the goal of AI independence in drug discovery.[4] Also, both Twist Bioscience and Genscript Biotech recently announced that they would partner with Eli Lilly’s AI-driven drug discovery ecosystem, TuneLab.[5]
Even without full AI-automation, biotechnology research is becoming increasingly AI-enabled, most clearly in early-stage discovery work (for example, in identifying new targets and designing small molecules, nucleic acids, and biologics).[6] Several AI-enabled therapies have already progressed to clinical trials,[7] with AI also being used to analyse clinical datasets to identify new biomarkers and targetable patient subgroups.
The problem with patenting AI-generated inventions Rights to ownership of AI output may be governed by both the AI provider’s contractual terms and IP law. Under the latter, a patent owner is typically the inventor(s), or a party that derives title to the invention from the inventor(s). Critically, in most jurisdictions, an inventor must be a human being,[8] meaning that a person or company cannot derive patent ownership from an AI ‘inventor’.
This does not mean that inventions involving the use of AI are inherently unpatentable. They can be, so long as AI is not the sole ‘inventor’. A human will be considered a (co)inventor if their contribution had a “material effect” on the final invention.[9] This may be the case where the human:
- Constructed inventive prompts to elicit a specific AI response;
- Analysed, refined, or altered the AI output in an inventive manner; or
- Trained an AI model to solve a particular problem.[10]
Merely doing “the initial prompt and the lab work” (as Anthropic attributed to its human scientists) may not be enough, where that lab work primarily verifies the AI output or reduces it to practice in a routine manner.
Lab notebooks may need to record AI input and output
While lab notebooks typically document experimental methods and results, they may not always detail the hypothesis generation process (for example, why certain drug targets, variants, or pathways are being tested) or data analysis workflows. Future inventorship disputes may require this type of evidence to demonstrate a human’s material contribution in identifying the problem to be solved, cleverly prompting the AI model, screening AI-generated candidates, or further modifying the AI output.
Records of those decisions (for example, through documentation of AI input and output and the model used) may therefore become critical in proving valid ownership over IP as we move closer towards AI-automated science.
[1] Claude discovers a novel enzyme system \ Anthropic
[2] EXCLUSIVE: Anthropic quietly sets up biology lab as it ramps AI drug program | Reuters
[3] Claude discovers a novel enzyme system \ Anthropic
[4] Roche outlines plans to move towards autonomous AI labs | Reuters
[5] Twist Bioscience | Twist Bioscience Joins Lilly TuneLab to Advance Antibody Drug Discovery; GenScript Joins Lilly TuneLab to Advance AI-Enabled Drug Discovery
[6] Riemer and Freund. Intelligent Pharmacy. 2026:4(2):145-158.
[7] For example, Borriello F, et al. J Infect Dis. 2026;jiag349; Xu Z, et al. Nat Med. 2025;31(8):2602-2610.
[8] Commissioner of Patents v Thaler [2022] FCAFC 62; Thaler v Vidal, 43 F.4th 1207 (Fed. Cir. 2022); J 8/20 and J 9/20 (Legal Board of Appeal, 2021/2022); Thaler v Commissioner of Patents [2023] NZHC 554; Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49.
[9] Polwood Pty Ltd v Foxworth Pty Ltd [2008] FCAFC 9 at [34].
[10] 7.2.8.5 Relevant Law – Entitlement/Inventorship | IPA Manuals; Federal Register :: Inventorship Guidance for AI-Assisted Inventions
