AI-designed drug turns back six aging clocks by up to 3 years

One company testing a lung disease treatment stumbled onto an unexpected number. Insilico Medicine, which designs new drugs using AI, published a paper in Nature Biotechnology on the 7th (local time) reporting that when it measured the blood of patients who had taken its AI-designed drug rentosertib for 12 weeks against six different aging clocks, their predicted biological age had gone down. The New York Times covered the study the same day, framing it as early data suggesting an AI-made drug could slow aging — and the story quickly caught fire.

The story starts last year. Insilico ran a Phase 2a trial of rentosertib in patients with idiopathic pulmonary fibrosis (IPF), and found a significant increase in the amount of air the lungs could hold. IPF is sometimes called "Alzheimer's of the lungs" — lung tissue thickens and scars, the lungs lose their ability to transfer oxygen into the blood, and the disease is chronic and can be fatal within a few years. The trial was originally designed simply to test whether the drug could help repair the lungs.

But Insilico built in one more layer from the start. CEO Alex Zhavoronkov told the New York Times that the drug was designed "also with the goal of extending patients' overall lifespan" — which is why, separately from the lung tests, the team measured proteins in patients' blood before and after dosing. According to the paper, of the 71 people enrolled in the trial, 42 consented to protein analysis, and researchers measured 2,841 proteins in their serum, then ran that data through six different aging clocks.

"Aging clock" is probably the least familiar term in this story, so think of it like the diagnostic equipment at an auto shop. The odometer tells you how many years a car has been driven, but a mechanic can listen to the engine and check the tire wear to separately judge how old the car actually is. An aging clock is an AI model that looks at protein levels in a person's blood to predict how old their cells and organs actually are biologically — not the age on their ID, but the age their body is telling.

The six clocks Insilico used are named ProtAge, two versions of OrganAge, PAC, ipfP3GPT, and PAOPAC — each reads proteins differently and produces a separate prediction. Yet over the 12-week trial, all six clocks showed the treated group's predicted age dropping below baseline, while the placebo group showed little change or a slight increase. Of 54 comparisons against placebo, 21 were statistically significant, concentrated especially at week 4 — and in the group taking 60mg once daily, four of the clocks calculated that patients had gotten 2.71 to 3.46 years younger by week 4. The fact that all six clocks pointed in the same direction is the basis for arguing this wasn't just chance.

Vadim Gladyshev, a Harvard Medical School professor who developed one of these clocks, told the New York Times this was "the first study to very clearly show that we can reduce predicted biological age." That should be read with the caveat that he's also a co-author on the paper, but language like this is rare in the aging-clock field.

The way the drug itself was made is also worth noting. Insilico has been one of the early companies pursuing neural-network-based drug discovery since its founding in 2014. Its first AI learns from thousands of medical records, blood tests, and papers to identify which proteins are driving a disease. Its second AI learns the physical shape and binding patterns of that protein and generates an entirely new molecule from scratch designed to attach to it and block its action. CEO Zhavoronkov likened this to "scanning a lock and then making a key to fit it." Rentosertib's lock is an enzyme called TNIK, and the molecular structure of the key that fits it was drawn not by a human, but by AI.

So AI shows up twice in this story: once in the AI that designed the drug, and once in the AI that measured its effects. The New York Times called this a milestone moment where the same class of technology that powers ChatGPT and Midjourney has made its way into drug labs outside the hospital. But to us, the measurement side is the more interesting story. Until now, testing whether an anti-aging drug actually worked meant watching people for decades to see if they lived longer. Aging clocks are a proposal to compress that decades-long wait into a 12-week blood test.

Of course, this result shouldn't be read as proof that an anti-aging drug has arrived. The sample is only around 40 people, and every subject in the trial had IPF, so nobody knows yet whether the same effect would show up in healthy people. Gladyshev himself acknowledged that the sample is small, that aging clocks aren't always reliable, and that it's far too early to draw conclusions. Eric Topol, a cardiologist and author of Super Agers, drew a similar line, saying the results "look encouraging, but there's no definitive trial yet to make a final call." Even getting rentosertib approved as an IPF treatment alone could still take years.

Even so, this paper is drawing attention less for its results than for its method. Evelyne Bischof, a longevity medicine professor at Tel Aviv University, said "these are methods we'll be using in future trials" — meaning that designing disease-drug trials to simultaneously measure aging markers from a few drops of blood could become the new standard going forward. That's also reflected in the paper's title, which ends by saying it supports concurrent anti-aging assessment.

We're likely to see more and more AI drug companies claiming their treatments also slow aging. When that happens, there are really only two things worth checking: which clock was used to measure it, and whether the same numbers hold up in healthy people, not just sick patients. Insilico answered the first question with six different clocks, and left the second one as homework for the entire industry. What is clear, at least, is that the work of turning aging into something measurable has begun.