The First AI-Designed Drug Just Worked in Patients. Now the Hard Question Begins

*A lung-disease pill that a machine helped invent improved patient breathing in a peer-reviewed trial. With 200-plus AI-found drugs now in human testing and Lilly betting $2.75 billion on the approach, the industry is about to learn whether algorithms can really shorten the road to the clinic.*

For two decades, "AI will revolutionize drug discovery" has been a pitch deck, not a result. That changed quietly when *Nature Medicine* published the Phase IIa data for rentosertib, a molecule that Hong Kong-based Insilico Medicine says was both discovered and designed using its generative-AI platform.

The headline number is small but real. In the 12-week study, patients with idiopathic pulmonary fibrosis (IPF) taking the 60 mg once-daily dose saw their lung capacity, measured as forced vital capacity, improve by an average of 98.4 mL. The placebo group declined by 20.3 mL over the same period. For a disease where the goal is usually to *slow* decline rather than reverse it, an actual gain in lung function is the kind of signal that makes pulmonologists look twice.

It is worth being precise about what this proves and what it does not. Rentosertib is a TNIK inhibitor, a target Insilico's software flagged as relevant to fibrosis. AI helped pick the target and shape the molecule; humans still ran the chemistry, the toxicology and the trial. Phase IIa is an early efficacy read on a modest number of patients, not a registrational study. No one has approved an AI-designed drug. What the data does establish is the first peer-reviewed, placebo-controlled clinical signal for a candidate that came out of a generative-AI pipeline, and that is the milestone the field has been waiting for.

Why this one matters more than the others

Rentosertib is not alone. By industry tallies, more than 200 drug candidates with AI in their discovery lineage are now in human trials, the bulk of them in early phases. The problem has always been attribution and outcomes: when a machine narrows a target list, it is hard to say the molecule that follows is meaningfully "AI's." A clean clinical readout cuts through that. It gives a concrete answer to the only question that matters to patients and investors alike, which is whether the candidate helps people.

The commercial signal arrived almost in parallel. In March 2026, Eli Lilly inked a deal worth up to $2.75 billion with Insilico, including $115 million up front, for exclusive licensing of select candidates plus access to its Pharma.AI platform. By Insilico's own count, it has put at least 28 generatively designed drugs into its pipeline, with roughly half at a clinical stage. Lilly is not buying a finished product; it is buying a discovery engine, and a pharma giant does not write that check on a press release.

The reality check

The skeptical case is still intact. AI has so far compressed the *earliest* stage of discovery, finding and shaping molecules, while the expensive, failure-prone middle of drug development remains stubbornly human and slow. Rentosertib must still survive larger, longer trials before anyone files for approval, and most Phase II successes never reach the market.

But the burden of proof has shifted. The question is no longer whether an AI-designed drug can reach a patient and do something measurable. It already has. The question now is how often, and how fast.

Fontes

• Rare Disease Advisor — AI-Discovered Drug Demonstrates Initial Safety and Tolerability in IPF
• Insilico Medicine — Nature Medicine Publication of Rentosertib Phase IIa Results
• EurekAlert! — Insilico announces Nature Medicine publication of Phase IIa rentosertib results
• CNBC — Eli Lilly reaches $2.75 billion deal with Insilico to bring AI-developed drugs to global market
• Fierce Biotech — Lilly inks R&D collab worth up to $2.75B with Insilico's AI engine
• Axis Intelligence — AI Drug Discovery 2026 Complete Analysis