Six Aging Clocks Moved Younger. Did an AI-Designed Drug Reverse Aging?

An experimental drug designed with the help of artificial intelligence has produced one of the most attention-grabbing findings in longevity research this year.

In a study published September 7 in Nature Biotechnology, researchers analyzed blood samples from people with idiopathic pulmonary fibrosis, or IPF, who had participated in a Phase 2a trial of rentosertib.

Rentosertib is an investigational drug designed to inhibit TNIK, a protein implicated in fibrosis and several biological processes associated with aging.

The researchers applied six different proteomic aging clocks to serum collected during the trial.

All six detected shifts toward a younger predicted biological age among participants receiving the drug.

That sounds like a human rejuvenation result.

It is not.

What the study found was a treatment-associated change in blood proteins that several computational models interpret as younger. It did not show that participants lived longer, became functionally younger, avoided age-related disease or experienced rejuvenation across their organs.

The distinction is not semantic.

It is the difference between an intriguing biomarker signal and evidence that an intervention has altered human aging.

What the researchers actually studied

The new analysis used samples from a previously published 12-week, randomized, double-blind, placebo-controlled Phase 2a trial of rentosertib in people with IPF.

IPF is a serious disease in which progressive scarring makes the lungs increasingly stiff and impairs their ability to function.

The original trial enrolled 71 participants across sites in China. Its primary endpoint concerned safety—not aging—and its clinical findings were preliminary. A secondary analysis suggested a possible improvement in forced vital capacity, a measure of how much air a person can exhale after taking a full breath, in one treatment group.

For the new proteomic study, researchers analyzed serial serum samples from 42 trial participants who completed the study, consented to the ancillary analysis and had measurements available at all four timepoints.

Thousands of circulating proteins were measured. The researchers then applied six published proteomic clocks—algorithms trained to estimate chronological age, mortality-related risk or other dimensions associated with biological aging.

Across the treated groups, the clocks generally moved toward younger estimates, while the placebo group showed little change or slightly older estimates.

The strongest agreement appeared around week four. In one treatment group, four clocks trained to estimate chronological age shifted approximately 2.7 to 3.5 years younger relative to baseline.

But that number requires context.

It is a change in a model-generated estimate, not evidence that the participants literally regained three years of life or reversed three years of whole-body aging.

A proteomic age is a prediction, not a physical age

Proteomic clocks look for patterns among proteins circulating in the blood.

Some proteins rise or fall predictably with age. Statistical models can combine those signals into an estimate that correlates with chronological age, disease risk or mortality across populations.

That can make the models scientifically useful.

If a clock estimates that one person has an older proteomic profile than another person of the same chronological age, that difference may contain information about health status or future risk.

But the output remains a prediction produced by an algorithm.

It is not a direct measurement of an underlying substance called “biological age.”

When an intervention changes a protein used by the model, the estimated age can change immediately. That may reflect a meaningful improvement in biology. It may also reflect treatment of a specific disease, suppression of inflammation, altered protein production or a direct pharmacological effect on the model's inputs.

A clock can register movement without proving that the organism has become younger in the broader sense people normally attach to that word.

The central problem: IPF and aging share biological signals

Separating disease improvement from aging modification is especially difficult in this study.

IPF is strongly associated with older age and involves fibrosis, inflammation, extracellular-matrix remodeling, cellular stress and senescence-related biology. Many of those processes also appear in studies of aging.

Treating IPF could therefore make an aging clock look younger simply because the disease-related protein pattern has improved.

The paper provides a particularly revealing example.

LTBP2, a protein involved in fibrotic signaling, was the only major contributor shared by all six clocks. Other influential proteins were also associated with fibrosis and extracellular-matrix remodeling.

In other words, the clocks were partly reading biology that the drug was specifically designed to change.

The authors recognized this problem. They wrote that proteomic clocks alone could not distinguish an anti-fibrotic effect from an effect on aging itself.

They performed additional analyses in an effort to separate the two.

Changes in lung function explained relatively little of the variation in the aging-clock results. The treatment also affected proteins and pathways associated with metabolism and cellular senescence, and one treatment pattern shifted age-associated proteins in the opposite direction from the trajectory observed with normal aging in UK Biobank data.

Those findings argue against the simplest explanation—that the clocks changed only because lung function improved.

But they do not eliminate disease-related confounding.

Lung function is only one dimension of IPF. A drug could change inflammation, fibrosis-related signaling or other aspects of the disease without those changes being fully captured by forced vital capacity.

Disease biology and aging biology are not cleanly separated systems. In this trial, they overlap by design.

Six clocks are more informative than one—but not six independent confirmations

Agreement among multiple clocks is a strength of the study.

The six models were built using different methods and included clocks trained for different purposes. A consistent direction across several models is harder to dismiss than a result from one proprietary score.

But “six clocks agreed” should not be interpreted as six independent clinical trials reaching the same conclusion.

The models were applied to the same 42 people, the same blood samples and overlapping biological information. Several of the chronological-age clocks were themselves highly correlated.

The statistical picture was also more qualified than the headline version.

The researchers evaluated multiple clocks, treatment groups and timepoints. Of 54 treatment-versus-placebo comparisons, 21 met the study's false-discovery threshold, with most of the signal concentrated at week four. The analyses used one-sided tests and an exploratory false-discovery-rate threshold of 0.10.

By week 12, fewer comparisons remained statistically significant, although the authors characterized the pattern as a plateau rather than a clear reversal of the earlier effect.

This is useful hypothesis-generating evidence.

It is not a definitive clinical demonstration of age reversal.

The people included in the analysis matter

The proteomic substudy was considerably smaller than the original trial.

Sixteen of the 71 randomized participants discontinued treatment before the 12-week study ended. The ancillary analysis included 42 people who completed the trial, consented to the proteomic assessment and had samples available at every required timepoint.

That means it was not a full intention-to-treat aging analysis of everyone who entered the trial.

The substudy included only 11 placebo participants, and the individual treatment groups contained between nine and 11 people.

All participants in the analysis were Asian, their average age was approximately 67 and every participant had IPF.

Those characteristics do not invalidate the result. They do limit how confidently it can be generalized—to other populations with IPF and especially to people without the disease.

The study was also short.

Twelve weeks can reveal whether a drug changes circulating proteins. It cannot establish whether the change persists, improves healthspan or translates into fewer age-related diseases years later.

What the study did not show

The study did not demonstrate that rentosertib:

• extended human lifespan;
• reduced all-cause mortality;
• prevented multiple age-related diseases;
• restored younger function across tissues or organs;
• improved strength, cognition, resilience or physical performance because of an anti-aging effect;
• produced the same clock changes in people without IPF; or
• is safe or beneficial for healthy people seeking longevity treatment.

Rentosertib remains an investigational drug being developed for a serious fibrotic lung disease. The study does not establish an anti-aging indication, and it provides no rationale for healthy consumers to seek or use the drug.

Several study authors are affiliated with Insilico Medicine, the company developing rentosertib, and the corresponding author is the company's founder and chief executive. The paper underwent peer review and included academic collaborators, but the commercial connection is relevant when interpreting an early result involving the developer's own product.

The AI story is real—but separate

Rentosertib is notable because AI tools contributed both to identifying TNIK as a potential therapeutic target and to designing the drug candidate.

That is a meaningful drug-development milestone.

But AI does not lower the evidence standard a drug must meet.

A molecule designed by an algorithm still has to demonstrate safety, efficacy and clinically meaningful benefit in rigorous human trials. A compelling origin story cannot substitute for those outcomes.

The original Phase 2a trial produced an encouraging lung-function signal, but it was small, short and designed primarily to examine safety. Larger and longer trials are needed to determine whether rentosertib meaningfully changes the course of IPF.

The aging claim requires an additional layer of evidence beyond that.

A drug could ultimately succeed as an IPF therapy without being a geroprotective drug. Conversely, a true effect on shared aging biology might exist even if a particular clinical endpoint in IPF proves disappointing.

Those are related questions, but they are not the same question.

Why the result still matters

None of these limitations makes the study unimportant.

The most valuable development may not be the claim that rentosertib made people younger. It may be the decision to collect longitudinal biological samples inside a controlled drug trial and examine whether a disease-targeted intervention also changes aging-associated systems.

Most purported anti-aging interventions are evaluated using uncontrolled before-and-after testing, small convenience samples or consumer assays with no placebo comparison.

This analysis had several advantages:

• participants came from a randomized, placebo-controlled trial;
• blood was collected repeatedly under a defined protocol;
• multiple published clocks were compared;
• individual proteins and biological pathways could be inspected;
• the clock signal was tested against lung-function changes and external population data; and
• the underlying analysis methods and much of the code and data were made available.

That is a much stronger scientific setting than purchasing a biological-age test, changing several behaviors at once and interpreting the second result as proof of rejuvenation.

The study offers a possible blueprint for including aging-related measurements in trials of treatments for diseases that share biology with aging.

Used carefully, those measurements could help researchers identify unexpected drug effects, compare mechanisms and decide which hypotheses deserve dedicated testing.

They should be treated as exploratory pharmacodynamic signals—not as surrogate proof that a therapy will extend healthy life.

A clinical-trial clock is not a consumer anti-aging test

The findings also should not be used to validate every commercially available biological-age score.

This study measured thousands of proteins using a research platform, collected samples at four standardized timepoints and compared changes across randomized treatment and placebo groups. Researchers could examine which proteins drove the results and whether those proteins aligned with known biological pathways.

A consumer receives an individual score without a randomized control group.

Changes in that score may be influenced by illness, inflammation, sample handling, analytical variation, recent behavior or the specific algorithm being used. Different clocks may respond differently because they were trained on different outcomes.

A trial-embedded biomarker can help researchers compare groups and generate hypotheses even when it is not sufficiently validated to direct decisions for an individual person.

Those are different use cases and should not be conflated.

What would stronger evidence look like?

A convincing case that rentosertib—or any intervention—modifies human aging would require more than a younger clock estimate.

Future studies would ideally include:

• aging endpoints specified before the analysis begins;
• substantially larger and more diverse populations;
• longer follow-up to determine whether the signal persists;
• replication by groups independent of the drug developer;
• comparison with other effective IPF treatments to help separate disease improvement from broader aging effects;
• measurements from additional biological layers or tissues;
• functional and clinical outcomes, not biomarkers alone; and
• evidence that biomarker changes predict lower disease burden, greater resilience or improved survival.

Testing in people without IPF could help separate fibrosis-related effects from broader aging biology. But an investigational drug should not be moved into healthy populations merely because a biomarker changed. Any such research would require an appropriate safety profile, ethical justification and formal clinical oversight.

The Modern Bio Take

Six proteomic clocks moving in the same direction is a signal worth investigating.

It is not proof that an AI-designed drug reversed human aging.

The study's most credible conclusion is narrower and more useful: a disease-targeted drug produced measurable changes in circulating proteins associated with aging, and those changes were detected across several computational models.

Some additional analyses suggest that the result may not be explained entirely by improved lung function. But fibrosis, inflammation and aging biology remain too entangled in this small IPF cohort to determine whether the drug produced genuine systemic rejuvenation.

This is how early science should be handled.

Do not dismiss an interesting signal because it is incomplete.

Do not promote an incomplete signal as an established outcome.

Measure it, test it again and ask whether the biomarker eventually connects to something people actually experience: better function, less disease and longer healthy lives.

Until then, the clocks moved.

We do not yet know whether human aging did.

Educational note: This article is for general educational purposes and is not medical advice. Rentosertib is an investigational drug and has not been established as an anti-aging treatment. Decisions about clinical-trial participation, testing, diagnosis or treatment remain the responsibility of the patient and their qualified healthcare practitioner.

Primary sources and editorial fact-check notes

• 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. Published September 7, 2026. DOI: 10.1038/s41587-026-03286-y. Primary source for the 42-person proteomic substudy, six-clock analysis, statistical methods, disease-versus-aging limitation, author affiliations and competing-interest disclosure.

• 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. Published June 3, 2025; 31:2602–2610. DOI: 10.1038/s41591-025-03743-2. Primary report of the 71-person randomized Phase 2a trial, including its safety endpoint, preliminary lung-function findings, discontinuations and study limitations.

• ClinicalTrials.gov record NCT05938920. Registration for the randomized Phase 2a rentosertib trial in idiopathic pulmonary fibrosis.