AI designed a lung drug — only humans can be its inventors

In 2023, biotech Insilico Medicine proudly announced that its generative AI platform had "discovered" a promising new molecule against pulmonary fibrosis. The candidate, then known as ISM001-055, targets idiopathic pulmonary fibrosis (IPF), a chronic disease that scars and stiffens lung tissue, affects roughly five million people worldwide, and leaves a median survival of just three to four years after diagnosis. The company's in-house Pharma.AI platform first used its PandaOmics engine to identify the kinase TNIK as a promising target, before its generative chemistry engine, Chemistry42, designed the compound itself — cutting the time from target identification to preclinical candidate down to 18 months. In March 2025, the USAN Council officially named the compound Rentosertib; in a Phase IIa trial in IPF patients, the highest dose produced a mean 98.4 milliliter improvement in lung function, against a 62.3 milliliter decline in the placebo group.

But when it came time to file the patent protecting that new chemical structure, Insilico made no mention of AI at all. Instead, the patent names five humans as "inventors," including CEO Alex Zhavoronkov, who leads the company he founded in 2014 at Johns Hopkins University and which has been headquartered in Hong Kong since 2019.

Only a human can be named on the patent

That gap between the press release and the legal filing is no accident — it's a direct consequence of US patent law. The statute, 35 U.S.C. § 100(f), defines an inventor as an "individual." In August 2022, in Thaler v. Vidal, an appeals court in Washington ruled that the term's plain meaning is a human being. The case was brought by patent attorney Ryan Abbott on behalf of Stephen Thaler, who wanted his AI system, DABUS, recognized as inventor of a food container whose intricate geometric surface improved heat transfer and stackability, arguing no human had contributed to the design. The court dismissed such "metaphysical matters" as beside the point: machines aren't people, so they can't be inventors. Case closed — with consequences that stretch far beyond food containers, since the same rule applies to any molecule an AI proposes.

"There needs to be a human inventor or there's no invention and no patent," says Sarah Korman, a patent attorney who is now chief business and legal officer at Isomorphic Labs, Alphabet's AI-drug-discovery spinout. Even the US Patent and Trademark Office (USPTO) acknowledges that an AI system, "like other tools, may perform acts that, if performed by a human, could constitute inventorship." The key question is therefore no longer whether an AI invented something, but whether a human contributed enough to the "conception" of the invention to count as a co-inventor, under the so-called Pannu factors.

Washington's about-face

How unsettled that standard is shows in the agency's own record. Under the Biden administration, the USPTO published a dedicated "Inventorship Guidance for AI-Assisted Inventions" in February 2024, meant to help companies determine when a human genuinely qualifies as a co-inventor of an AI-assisted discovery. Since Trump took office, the agency has reversed course: AI is now officially treated as just a tool, like a calculator, that doesn't even need to be mentioned. Between the detailed 2024 guidance and today's "calculator" stance lies a significant shift that leaves companies without a reliable compass — lawyers nonetheless keep recommending the stricter documentation practice, because Abbott warns that a granted patent can later be invalidated if it's shown to list the wrong inventors.

• Training-data curation: which biological datasets were chosen, normalized, or excluded
• Reward-function design: how binding affinity, synthetic accessibility, and toxicity risk were weighted
• Candidate selection: why one molecule was chosen from the AI-generated shortlist
• Experimental validation: which synthesis routes and animal studies humans then decided on

Human chemists still have to synthesize the drugs, create variants, and test them on animals. That's the person who is going to be named on the patent. And even if you decided to completely roboticize this process, including the experiments, someone will still push the button and give the budget.

What's at stake for investment — and beyond the US

The stakes reach well past a single patent dispute. Over a drug's commercial life, a compound patent typically accounts for 60 to 80 percent of its net present value — if inventorship is successfully challenged, the economic core of the investment goes with it. Should US policy keep AI-generated outputs permanently outside reliable patent protection, Abbott warns, that could dampen investment in AI-driven drug development altogether — much as the US Copyright Office already refuses copyright protection to images and text generated entirely by AI, a stance that has alarmed groups like the Motion Picture Association of America. The European Patent Office, along with the UK, Chinese, and Japanese patent offices, also require a human inventor — but their standards diverge in the details, so a single international PCT filing doesn't resolve the discrepancies between jurisdictions.

For biotech and pharma companies leaning on generative AI for drug discovery, the takeaway is concrete: securing internationally enforceable patents now means documenting every human decision — from data curation to lab validation — regardless of how loosely the USPTO currently interprets that duty. The UK's Intellectual Property Office applies the same human-inventor requirement as the USPTO, so the paper trail matters just as much for a filing out of London or Cambridge as one out of Boston. Investors backing an AI-native biotech are increasingly likely to scrutinize that documentation as closely as the clinical data, since a patent invalidated after the fact can wipe out most of a company's value.