AI Designed a Drug That Made Six Ageing Clocks Run Backwards

An AI-discovered, AI-designed drug lowered predicted biological age across six independent ageing clocks in a phase 2a trial. Here is what that does and does not prove.

For years, "reversing biological age" has been the kind of phrase that belongs in a supplement advert, not a peer-reviewed journal. On 7 September 2026, that line blurred. Nature Biotechnology published an analysis showing that a drug called rentosertib lowered the predicted biological age of patients in a clinical trial, and that six separate ageing "clocks", built by different laboratories using different methods, all agreed on the direction. The drug is unusual for a second reason: both its biological target and its chemical structure were found and designed by artificial intelligence. It is a real advance. It is also a neat lesson in how easily a careful result gets oversold.

The company behind the work, Insilico Medicine, did not run a fresh anti-ageing trial. Its team and collaborators at Harvard, Stanford, the Broad Institute and Peking University went back to blood samples already banked from a completed lung-disease study.

That earlier study was a phase 2a trial of rentosertib in idiopathic pulmonary fibrosis (IPF), a progressive scarring of the lungs. It randomised 71 patients to one of three doses or a placebo for 12 weeks. For the new analysis, the researchers took the 42 participants who had complete samples and consent, and measured 2,841 proteins in their blood using a platform called Olink.

They then fed those readings into six independently built proteomic ageing clocks: ProtAge, two versions of OrganAge, PAC, ipfP3GPT and PAOPAC. Despite being trained on different data with different goals, all six pointed the same way. Patients on rentosertib looked biologically younger than those on placebo. The effect peaked at week four. In the strongest dose groups it came to roughly three to four years of predicted biological age, with one clock suggesting up to six. Placebo patients showed no comparable shift.

The science you need: what an ageing clock actually measures#

An ageing clock is a statistical model, usually built with machine learning, that estimates a person's "biological age" from molecular data rather than their birthday. The premise: two 65-year-olds can age at different rates, and the body leaves chemical clues about which is which.

Early clocks read chemical tags on DNA. The clocks used here read the proteome, the thousands of proteins circulating in blood, which reflect inflammation, metabolism, tissue repair and cellular stress in close to real time. A proteomic clock is trained on samples from large populations: the model learns which protein patterns track with chronological age, or with the risk of dying, and can then score a new blood sample. Some tools go further and estimate the age of individual organs, such as the heart or lungs, from the same draw.

The second idea is AI drug discovery. Rentosertib did not begin with a chemist's hunch. Insilico's software first flagged a protein called TNIK as a promising target, a molecular switch worth blocking, because it turned up in six recognised "hallmarks of ageing", the core processes thought to drive age-related decline. A separate generative system, the kind that designs novel molecules, then drew the drug to fit that target. Insilico says the journey from picking the target to naming a preclinical candidate took about 18 months, fast by industry standards. This is why the study is billed as a first. Earlier "longevity" experiments mostly repurposed old drugs such as rapamycin or metformin, whereas both the target and the molecule here came from AI.

Why this matters#

The headline is not really "a drug reversed ageing". The more durable point is about method.

The agreement between clocks is the interesting part. Any single ageing clock can be fooled, since a flattering number might just reflect its training data. Getting six clocks that share neither their features nor their training sets to move together is harder to wave away. As Nobel laureate Michael Levitt put it, what convinces him is not the size of the effect but the agreement.

The signal also did not look like a simple by-product of treating the lung disease. The dose that helped breathing most was not the dose that shifted the clocks most, which hints that the drug touches ageing biology somewhat separately from its effect on the lungs. Looking at the proteins themselves, the team reported that rentosertib behaved as a senomorphic: it quietened the harmful signals released by worn-out "senescent" cells rather than killing those cells outright, and it dialled down growth-signalling pathways linked to faster ageing.

Most practically, the study is a template. Most disease trials already bank blood. The proposal is to run ageing clocks on those samples as an exploratory extra, so a drug's effect on ageing biology can be spotted inside an ordinary disease trial instead of waiting years for a dedicated longevity study. If regulators eventually accept such biomarkers, geroprotective candidates could surface far earlier. For a field where "ageing" is not yet an approved thing to treat, that is a plausible way in.

Read the asterisks#

The authors are refreshingly candid about the limits, and it is worth taking them at their word.

The sample is small, and everyone in it was ill. Forty-two people, all with a serious lung disease, is a discovery-scale dataset, not proof. The central ambiguity is unresolved: did rentosertib slow ageing, or did it simply make the blood of IPF patients look younger as their disease eased? Treating almost any illness can nudge inflammatory and metabolic proteins in a "younger" direction. The clocks cannot separate those explanations on their own, which is why the cleanest next test, as Levitt noted, is healthy volunteers.

The signal was also short-lived here: it peaked at week four and faded to a plateau by week 12, with fewer comparisons staying significant. A convincing anti-ageing effect ought to persist. Some supporting pathway analysis used a relaxed statistical threshold that tolerates more false positives than confirmatory studies accept, and the analysis was retrospective, with ageing endpoints chosen after the trial rather than before.

Then there is the biomarker trap. A younger protein profile is not the same as a longer, healthier life. Regulators draw a firm line between a biomarker, meaning any biological signal, and a validated surrogate endpoint, meaning a signal proven to stand in reliably for how a patient feels, functions or survives. No proteomic ageing clock has crossed that line. The study did not measure mortality or physical function, and it did not check the result against a DNA-based clock for comparison. It is also company-led, with Insilico's chief executive as first author. That is not disqualifying, but it is a reason to want independent replication. Helpfully, the data and code have been made public.

Finally, timelines. Rentosertib still has to prove itself as a lung drug against established antifibrotics such as nintedanib and pirfenidone, and it has only just entered phase 3. Any credible "longevity" claim sits several trials and several years away.

Expert perspective: how this compares#

Set against the field, the work is incremental in some respects and genuinely new in others. Proteomic clocks are not new. Large studies have already shown that blood-protein signatures predict mortality and disease, and that organ-specific clocks can flag which body systems are ageing fastest. AI-assisted drug discovery is not new either; rentosertib's earlier chapters were published in Nature Biotechnology in 2024 and Nature Medicine in 2025.

What is different is the join. Previous geroscience trials, such as those on metformin, leaned on repurposed generics. Here, an AI-nominated target thought to sit at the crossroads of ageing and disease was matched with an AI-designed molecule, then tracked with a whole panel of ageing clocks in one cohort. That traceability, from a computational hypothesis through to human protein data, is what stands out. It is also what lets outside scientists test the original idea instead of taking a press release on trust.

Key takeaways#

1. An AI-discovered, AI-designed drug, rentosertib, lowered predicted biological age across six independent proteomic ageing clocks in a 42-patient phase 2a subset, published in Nature Biotechnology.
2. The strength of the finding is the agreement between clocks that share no training data, not the raw number of years.
3. It is a biomarker signal, not proof that anyone aged backwards. The effect peaked at week four, and every patient was being treated for a serious lung disease.
4. The lasting contribution may be the blueprint: measuring ageing biology inside conventional disease trials.
5. The decisive tests are still to come, namely phase 3 results, a pre-planned biomarker analysis, and ideally data from outside lung disease.

Frequently asked questions#

Did this drug reverse human ageing? No. It shifted blood-protein patterns that ageing clocks read as "younger". That is a promising signal, not evidence of longer life or restored health.

What is rentosertib? A small molecule that blocks a protein called TNIK. It was discovered and designed with AI and is being developed for idiopathic pulmonary fibrosis, a scarring lung disease.

Why test an anti-ageing idea in lung-disease patients? Because the samples already existed, and because IPF is an age-related disease that shares biology with ageing. That overlap is also the study's main interpretation problem.

Why use six clocks instead of one? A single clock can produce a flattering result for its own reasons. Agreement across six built by different groups makes a model-specific fluke less likely, though they did all analyse the same 42 people.

Can I take this drug to slow ageing? No. It is an experimental therapy in trials for a specific lung disease, with no ageing indication. Nothing here is medical advice.

What would make the result convincing? Sustained effects in a larger, longer, pre-registered study; confirmation in healthy or non-IPF participants; and links to real outcomes such as physical function or survival.

Glossary#

Ageing clock: a machine-learning model that estimates "biological age" from molecular data such as proteins or DNA marks.

Proteome / proteomic: the full set of proteins in a sample; proteomic clocks read blood proteins.

Biological age: an estimate of how old the body seems physiologically, which can differ from age in years.

TNIK: the protein rentosertib blocks. AI flagged it as relevant to both fibrosis and ageing.

Senomorphic: a drug that suppresses the harmful signals from worn-out ("senescent") cells without killing them. A senolytic, by contrast, removes them.

Hallmarks of ageing: a widely used list of the core biological processes thought to drive ageing.

Biomarker vs surrogate endpoint: a biomarker is any measurable biological signal; a surrogate endpoint is a biomarker proven reliable enough to stand in for real clinical benefit. Ageing clocks are the former, not the latter.

Idiopathic pulmonary fibrosis (IPF): a progressive, incurable scarring of the lungs affecting roughly 5 million people worldwide.

References#

1. Zhavoronkov, A., Galkin, F., Chen, S. et al. "Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment." Nature Biotechnology (2026). https://doi.org/10.1038/s41587-026-03286-y (peer-reviewed primary source).
2. Insilico Medicine. "Insilico's AI-Driven IPF Candidate Rentosertib Shows Potential for Biological Age Reversal, as Assessed by Six Proteomic Aging Clocks." Press release, 7 September 2026. https://insilico.com/news/rnt0709261-rentosertib-proteomic-aging-clocks
3. Xu, Z., Ren, F., Wang, P. et al. "A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial." Nature Medicine 31, 2602-2610 (2025). https://doi.org/10.1038/s41591-025-03743-2
4. Ren, F., Aliper, A., Chen, J. et al. "A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models." Nature Biotechnology 43, 63-75 (2025). https://doi.org/10.1038/s41587-024-02143-0
5. Pun, F. W. et al. "Hallmarks of aging-based dual-purpose disease and age-associated targets predicted using PandaOmics." Aging (Albany NY) 14, 2475-2506 (2022). https://doi.org/10.18632/aging.203960
6. Argentieri, M. A. et al. "Proteomic aging clock predicts mortality and risk of common age-related diseases in diverse populations." Nature Medicine (2024). https://www.nature.com/articles/s41591-024-03164-7
7. Oh, H. S.-H. et al. "Organ aging signatures in the plasma proteome track health and disease." Nature 624, 164-172 (2023). https://www.nature.com/articles/s41586-023-06802-1
8. "Plasma proteomics links brain and immune system aging with healthspan and longevity." Nature Medicine (2025). https://www.nature.com/articles/s41591-025-03798-1
9. Johnson, O. "Insilico Rentosertib Aging Study Finds Younger Biomarkers, Not Younger Patients." remio, 8 September 2026. https://www.remio.ai/post/insilico-rentosertib-aging-study-finds-younger-biomarkers-not-younger-patients (independent analysis).
10. "AI-Discovered Drug Reverses Aging Markers in Study, Biotech Says." Bloomberg, 7 September 2026. https://www.bloomberg.com/news/articles/2026-09-07/ai-discovered-drug-reverses-aging-markers-in-study-biotech-says
11. "AI-designed drug candidate reverses biological age in clinical study." News-Medical, 7 September 2026. https://www.news-medical.net/news/20260907/AI-designed-drug-candidate-reverses-biological-age-in-clinical-study.aspx