A drug designed by AI just worked in people
Not in a simulation. In a Phase IIa trial, in patients with idiopathic pulmonary fibrosis — a disease that scars the lungs and, for most people who get it, shortens life to a handful of years.
The compound is ISM001-055, from Insilico Medicine. Both halves of the claim matter: the biological target was found by AI, and the molecule aimed at it was designed by AI. The company reports the route from project start to preclinical candidate took over 60% less time than the industry norm. Eli Lilly has since signed a deal with Insilico worth up to $2.75 billion for worldwide rights to candidates from the platform. Separately this year, MIT researchers used generative AI to design new antibiotics that cured drug-resistant infections in mice.
Most of what this industry calls a breakthrough is a demo. This is not a demo. Phase IIa is the stage where enthusiasm usually dies, because it is the first time a drug has to do something measurable in a human being rather than a model.
The honest caveat: Phase IIa is not approval, and plenty of drugs that pass it fail later. But the claim being tested was never "AI cures disease". It was narrower and much harder — can a machine choose a target worth chasing, and design something that hits it?
On this evidence: once. And once is a number that used to be zero.
Anyone who has watched a family member wait for a treatment that does not exist yet.
Phase III, which is longer, larger and unforgiving. That is the one that decides.
Somewhere there is a person with scarred lungs who may one day be handed a medicine that no human being first imagined. Whatever else gets argued about this technology this year, that part happened.
Sources: GlobeNewswire bioinformatics report, August 6, 2026; AI World Journal; News-Medical.