Gangsta AI

An AI Designed a Drug From Scratch. Six Aging Clocks Say It Made Patients Younger.

Breathtaking

By Keanu Reeves · 2026-09-07 · 4 min read

A machine picked the target. A machine drew the molecule. And this week, six aging clocks that were never built to talk to each other all looked at the blood of the people who took it and said the same quiet, astonishing thing: they got younger.

The drug is rentosertib. On September 7, in *Nature Biotechnology*, researchers from Insilico Medicine and academic labs reported that a small-molecule TNIK inhibitor — dreamed up almost entirely by AI — appeared to reverse biological age in patients with idiopathic pulmonary fibrosis, a brutal, scarring lung disease with few good options.

What the machines actually did

Most drugs start with a human hunch. This one didn't. Insilico's PandaOmics platform combed through mountains of biology and flagged TNIK — a gene tangled up in *six* of the recognized hallmarks of aging — as the thing to hit. Then its generative chemistry system, Chemistry42, designed the actual molecule to hit it. That combination is the milestone: rentosertib is described as the first drug candidate with both an AI-discovered target and an AI-designed molecule.

“A machine chose the enemy. A machine forged the weapon. Humans ran the trial. That's new.”

Why six clocks matter

Here's the part I keep turning over. The team took 12-week blood samples from 42 patients in a randomized, double-blind, placebo-controlled Phase IIa trial and ran them through six different proteomic aging clocks — independent models developed by separate groups at Harvard, Oxford, Peking University and Insilico itself.

Nobody trusted a single clock. They didn't have to. All six independently pointed the same direction: patients on rentosertib looked biologically younger than they should have. When six models that were built by rival labs, on different math, all agree — that's not one algorithm's opinion. That's a consensus. The drug has now advanced to a Phase III trial.

The lesson hiding in the method

That's the beautiful part, and it's also the whole point. The scientists didn't ask one model and believe it. They asked *many*, built by *different people*, and only leaned on the answer where the models converged.

We live in that same moment now. Every week another frontier model — ChatGPT, Claude, Gemini, Grok, and a dozen more — leapfrogs the last and claims the crown. No single one is reliably best at everything, and any single one can be confidently wrong. So when the question actually matters, the smart move is the one those aging clocks made: don't gamble on one voice.

That's the idea behind Gangsta AI — ask the same question across 30+ leading models at once, then get one cross-checked, cited verdict instead of betting your answer on a single roll of the dice. Consensus beats a lucky guess. Six clocks just proved it in a lung-disease ward.

A drug no human fully invented may be turning back the clock in real patients. However you feel about that, take a breath. It's kind of breathtaking. And the way they proved it — by refusing to trust one model alone — is a lesson the rest of us can use today.

Want to see how the top models line up when the answer matters? Start with our best AI models rundown.

Try Gangsta AI free →

More: Best AI models · Compare all AI · Frontier Models · All articles