Overview
Insilico Medicine’s new AI-designed drug rentosertib, now entering Phase III trials, raises interesting inventorship questions. According to Insilico’s press release, Rentosertib’s “target was identified with AI” and its “chemical structure was designed with generative AI.” Other published sources refer to the molecule as AI “discovered” and “generated”. Yet Insilico’s two patents on that drug name only human inventors.
Inventorship is assessed claim by claim. For a jointly invented claim, that inventorship is defined as “the individuals collectively who invented or discovered the subject matter of the invention.” 35 U.S.C. § 100(f).
In 2022, Thaler v. Vidal held that inventors are limited to natural persons. But Thaler did not decide whether inventions partially conceived by AI can be patented. Instead, it noted that it was not confronted with “the question of whether inventions made by human beings with the assistance of AI are eligible for patent protection.” The Patent Office’s November 2025 guidance says there is “no separate or modified standard for AI-assisted inventions.” It calls AI systems “instruments used by human inventors.” It asks whether the natural person “possessed knowledge of all the limitations of the claimed invention” in a form definite enough that ordinary skill could reduce it to practice.
Insilico’s patents comply with Thaler’s natural person limitation on their face. No doubt, humans had some role in the conception of at least some aspects of the inventions claimed in those patents. But Insilico’s innovative use of AI in the discovery process is well documented and points to a substantial role by AI as well.
There appear to be two AI systems central to Insilico’s work: PandaOmics, which hunts disease targets, and Chemistry42, which designs molecules. A Nature Biotechnology paper on the discovery, authored by Insilico scientists and inventors, says that the drug’s target, TNIK, “was identified by tasking the PandaOmics target-discovery platform to search for new anti-fibrotic targets.” TNIK is a kinase, a kind of signaling protein, implicated in fibrosis.
Among the candidates, “TNIK scored as number 1” on the platform’s kinase setting. For the molecule, “Chemistry42 first applied 30 generative models in parallel to generate compound structures.” Of what came back, rentosertib “showed the most desirable drug properties with strong selectivity for inhibiting TNIK.” The paper describes “roughly 18 months from target discovery to preclinical candidate nomination” for the whole program.
In addition to PandaOmics and Chemistry42, humans had a significant role too. The same Nature Biotechnology paper records the choices they made. It says the authors “selected the ATP-binding site as a pocket for compound generation.” They further configured the platform to produce small-molecule structures that bind at that site, and applied an additional hydrophobic pharmacophore to narrow what it returned.
Insilico’s general counsel, Joanna Wang, provided further detailed descriptions of Insilico’s discovery process in the Journal of Law and the Biosciences. The platform’s output, she writes, “was not a single compound, but rather a diverse set of candidate molecules that were then prioritized based on multiple criteria.” Choosing among them “required significant human expertise to interpret the results and select the most promising candidates for synthesis and testing.” Wang also describes an Insilico documentation process meant to capture each named inventor’s contribution to conception.
To evaluate conception, it is necessary to look at Insilico’s patent claims. U.S. Patent No. 11,795,160 (the “’160 patent”) claims a genus of compounds that includes rentosertib. U.S. Patent No. 11,739,078 (the “’078 patent”) claims methods of inhibiting a kinase using a similar genus including rentosertib.
A challenger might argue from Insilico’s own account that AI conceived at least the claimed molecule (rentosertib), so the humans named on the patents were not, collectively, those “who invented or discovered the subject matter of the invention.” Insilico might respond that it followed Thaler to a T, naming all humans who participated in conception.
This post does not express a view on whether Insilico’s patents have an inventorship problem. But it is interesting to consider the remedy if a court were to find human inventorship lacking. Section 256 lets a patentee correct a mistaken inventor list. But suppose the one who did the conceiving is the AI itself. Thaler forecloses naming it, and there is no human to add in its place, so the error could not be corrected. That would be a live validity issue. In April the Federal Circuit held in Fortress Iron, LP v. Digger Specialties, Inc. that “a patent which incorrectly lists its inventor(s) and cannot be corrected according to law is invalid.”
No court or agency has yet said how much of a molecule’s conception must happen in a person’s mind. For a patent on the wrong side of the eventual line, there may be no cure. Insilico is not waiting to be asked. Its scientists’ role is documented in the journals, and its general counsel has published the company’s account. The answer, when it comes, will find the record ready.
The Weekly Inference is a recurring Step Into IP feature. Nothing here is legal advice.
