In January I wrote that drugs designed by artificial intelligence had finally entered the exam room, and that the honest scoreboard for 2026 read: molecules in the room, no grades posted. Nine months later the scoreboard has a new line. On September 10, Insilico Medicine announced that the first patient had been dosed in GENESIS-IPF-3, a Phase III trial of rentosertib, a drug for a lung disease called idiopathic pulmonary fibrosis. The AI-drug company calls it the world's first Phase III trial of a generative-AI-driven drug.
"First" is where hype likes to live, so let me unpack exactly what that sentence earns.
The Disease and the Drug
Idiopathic pulmonary fibrosis is a disease in which lung tissue scars and stiffens; "idiopathic" is medicine's word for "cause unknown." Insilico puts the number of people living with it at about five million worldwide, with a median survival of three to four years.
Rentosertib works by blocking a protein called TNIK. What makes it unusual on my scoreboard is that, according to Insilico, AI did two jobs here: it picked the target, and it generated the molecule built to hit it. The trial's lead investigator, Professor Zuojun Xu of Peking Union Medical College Hospital, put the first half plainly: "TNIK, the target driven by AI, had never previously been linked to fibrosis."
Recall my January markers. One of the signals that would move this desk's scoreboard up was AI finding targets the field had overlooked. Rentosertib is a candidate for exactly that. A candidate is all it is.
What Earned the Promotion
The evidence that got rentosertib this far is a Phase IIa study published in Nature Medicine in June 2025: 71 patients at 22 sites in China, split among a placebo and three dosing schedules, for 12 weeks. Its primary goal was safety, and it met that goal. On lung function, the highest-dose group gained an average of 98.4 milliliters of forced vital capacity, roughly how much air you can blow out in one hard breath, while the placebo group lost 20.3 milliliters.
That is an encouraging signal, and a small one. Seventy-one patients across four groups works out to fewer than twenty per group on average, measured for three months, against a disease that plays out over years. Small trials produce dramatic numbers easily. That is precisely why Phase III exists.
The Exam Itself
GENESIS-IPF-3 is the real test: about 320 patients at 47 centers in China, randomized and double-blind, against placebo, for 52 weeks. The primary endpoint is the yearly rate of decline in lung capacity. No more twelve-week snapshots. A full year, a large group, and a control arm with the power to embarrass the drug.
History says to expect embarrassment more often than not. The biotech industry's own large analysis of development programs from 2006 to 2015 found that fewer than one in ten candidates entering human trials went on to win approval, and that roughly 70 percent stalled at Phase II. Rentosertib has cleared the stage where most drugs die. It has not cleared the last one.
How I'll Keep Score
Three markers, posted in advance as usual.
First, the clock. Xu estimates about three to four years from the start of Phase III to approval, and that is under favorable conditions. There will be no verdict this school year, and anyone declaring one is selling something.
Second, the map. Every one of the 47 centers is in China. That is not a flaw in the science, but I will be watching whether regulators elsewhere treat the results as enough or ask for trials of their own.
Third, and most important: one drug is an anecdote, not a portfolio. My January test was whether AI-designed candidates, as a group, fail less often than the industry's dismal average. A clean win for rentosertib would prove that one molecule works, not that the method beats human chemists. It would, however, be the first grade ever posted in that column.
The exam has started. It runs fifty-two weeks per patient, and this desk will be here when the proctor collects the papers.
