AI-Designed Longevity Drugs Are in Human Trials. Do Any of Them Work?

A lung drug, six clocks and a very large claim#

In September 2026, Nature Biotechnology published a paper with an unusually modest title for what it implied. Researchers took blood from people with a scarring lung disease who had swallowed an experimental pill for 12 weeks, ran the samples through six different "ageing clocks", and found that all six predicted a lower biological age in the treated groups (Zhavoronkov et al., 2026). The pill, rentosertib, was designed with generative AI. Its target was picked by AI too (Ren et al., 2024).

That combination is catnip for headlines. "AI drug reverses ageing" practically writes itself. It is also, as the authors themselves say, not what the data show. The paper states plainly that proteomic clocks "cannot deconvolute aging- and disease-specific effects" (Zhavoronkov et al., 2026). In other words, a lung that is less inflamed may simply make your blood look younger.

Still, something real is happening. For the first time, drugs that were found or shaped by machine learning, and that are explicitly pitched at the biology of ageing, are being tested in people. Some have early human data. One has started a Phase 3 trial. So this is a good moment to sort the evidence from the excitement, which is what the rest of this piece tries to do.

What you need to know first: ageing biology, AI and why trials are tricky#

Geroscience is the idea that ageing itself drives most chronic disease, so slowing the underlying biology might delay many diseases at once. Researchers usually describe that biology as a set of "hallmarks". The current list has 12, including genomic instability, loss of protein quality control, disabled autophagy (the cell's recycling system), cellular senescence and chronic inflammation (López-Otín et al., 2023).

A few terms come up again and again:

• Senescent cells are damaged cells that stop dividing but refuse to die. They leak inflammatory signals known as the SASP (senescence-associated secretory phenotype). People sometimes call them "zombie cells".
• Senolytics are drugs that kill senescent cells. Senomorphics leave the cells alive but quieten their harmful signalling.
• Biological age is an estimate of how old your body "looks" at the molecular level, as opposed to your birthday age. Ageing clocks are statistical or machine learning models that produce that estimate from measurements such as DNA methylation or blood proteins. A widely used proteomic clock, ProtAge, was trained on blood protein data from the UK Biobank and predicts mortality and several age-related diseases (Argentieri et al., 2024).
• AI drug discovery covers two different jobs. One is target discovery: sifting genetic, protein and clinical data to find which molecule in the body a drug should hit. The other is molecule design: generating and refining chemical structures that bind that target and behave like a medicine.

Here is the catch for longevity companies. Regulators approve drugs for diseases, and "ageing" is not on the list. A trial also cannot easily wait 20 years to see whether people live longer. The usual workaround is to test a drug in an age-related disease first, then look for signs that it touches ageing biology along the way. The rentosertib paper describes this explicitly as a "dual-purpose" trial design (Zhavoronkov et al., 2026).

Rentosertib: from AI target to Phase 3#

How it was found#

Rentosertib (earlier code names INS018_055 and ISM001-055) comes from Insilico Medicine. The company's AI tools flagged an enzyme called TNIK (TRAF2- and NCK-interacting kinase) as a target for fibrosis, the stiff scar tissue that builds up in organs. Its generative chemistry tools then proposed molecules to block it. The team reports that the work took about 18 months from target discovery to a nominated preclinical candidate (Ren et al., 2024). That is quick by industry standards, though "fast" is not the same as "right".

A Phase 1 trial in 78 healthy volunteers found acceptable safety and drug levels (Ren et al., 2024).

What Phase 2a showed#

The Phase 2a trial enrolled 71 people with idiopathic pulmonary fibrosis (IPF), a progressive scarring disease of the lungs that mainly affects older adults. Participants received one of three doses or a placebo for 12 weeks (Xu et al., 2025). The main goal was safety. Side effects were similar across groups: 72% to 83% of people on the drug had at least one adverse event, against 71% on placebo. The most common reasons for stopping treatment were liver toxicity or diarrhoea (Xu et al., 2025).

On lung function, people on the highest dose (60 mg once a day) gained an average of 98.4 ml in forced vital capacity (FVC, the amount of air you can forcibly breathe out), while the placebo group lost 20.3 ml (Xu et al., 2025). That sounds impressive, and it is encouraging. But FVC was a secondary endpoint, the groups had 17 or 18 people each, and the confidence intervals are wide. A 2026 review was careful to call the lung findings "secondary and exploratory" (Agboola et al., 2026).

Figure: Lung function change in the rentosertib Phase 2a trial. The overlapping error bars are a reminder of how small the groups were (Xu et al., 2025).

An independent network meta-analysis, which compares many IPF drugs indirectly, ranked rentosertib among the options most likely to preserve FVC. The authors rated the certainty of that evidence as low (Xu J. et al., 2026).

The Phase 3 test#

Insilico has now moved into a 52-week Phase 3 trial called GENESIS-IPF-3, planned for about 320 participants at 47 centres in China. The primary endpoint is the yearly rate of FVC decline, and the first patient was dosed in September 2026 (Insilico Medicine, 2026). Company timelines quoted in trade press point to initial results around the end of 2029 (Bio-IT World, 2026). Note what this trial measures: lungs. It is not designed to prove anything about ageing.

The ageing-clock result, read carefully#

What the paper found#

The Nature Biotechnology analysis used serum from 42 of the Phase 2a participants and tracked roughly 2,841 blood proteins (Longevity.Technology, 2026). The six clocks were ProtAge, two versions of OrganAge, PAC, PAOPAC and Insilico's own ipfP3GPT. All six predicted lower biological age in treated arms (Zhavoronkov et al., 2026). The most consistent signal came at week 4 in the 30 mg twice-daily group, where five of six clocks moved in the "younger" direction (Longevity.Technology, 2026).

Two details are interesting. First, the dose that best improved lung function (60 mg once daily) was not the one that moved the clocks most. Second, changes in FVC explained very little of the change in predicted age, with a median R² of 0.06 (Longevity.Technology, 2026). Pathway analysis also pointed to shifts in senescence and metabolic processes alongside the anti-fibrotic effect (Zhavoronkov et al., 2026). That is at least a hint that something beyond "the lungs got better" may be going on.

Why caution is warranted#

I find the agreement between clocks more persuasive than any single number. But several things temper my enthusiasm. The sample is 42 people over 12 weeks. The analysis was secondary, run after the trial. And independence is thinner than it looks: by Longevity.Technology's reading, four of the five clock-development teams are co-authors, and only PAC comes from a fully independent group (Longevity.Technology, 2026). The author list itself is led by Insilico staff (Zhavoronkov et al., 2026).

The big confounder remains IPF. It is an inflammatory disease that leaves fingerprints in blood proteins, so treating it could shift the very markers that clocks read. The authors say as much (Zhavoronkov et al., 2026). A clean test would need a population without lung disease, longer follow-up and direct measures of senescent cells. Until then, "lower biological age" is a hypothesis with a promising first data point, not a finding.

Beyond rentosertib: who else is in the clinic?#

Rentosertib is the furthest along, but it is not alone. The table below sorts the most visible programmes by how AI or large-scale data analysis was used and how far each has reached in people.

BioAge: start with long-lived humans#

BioAge Labs takes a different route from Insilico. Instead of generating new chemistry with AI, it mines data from human ageing cohorts to choose targets. Its current lead, BGE-102, blocks NLRP3, a protein complex that switches on inflammation; the company says its cohort analyses linked lower NLRP3 activity with longer life (BioAge, 2026). In April 2026, it reported Phase 1 data in people with obesity and high levels of hsCRP, a blood marker of inflammation. Median hsCRP fell by 80% to 86% within days, and no serious adverse events were reported (BioAge, 2026). A cardiovascular-risk Phase 2 trial is planned.

BioAge's history is a useful corrective to platform optimism. In December 2024 it halted a Phase 2 obesity trial of its previous lead, azelaprag, after 11 of 204 participants on the drug showed raised liver enzymes (BioAge, 2024). Good targets can still come with bad surprises.

Rubedo: an AI-found senolytic for skin#

Rubedo Life Sciences says its ALEMBIC platform uses AI to find targets inside pathological senescent cells. Its lead, RLS-1496, is a cream that modulates GPX4, an enzyme involved in a form of cell death (Rubedo, 2025). Preliminary Phase 1 results in psoriasis reported fewer senescent cells at mid and high doses and about a 20% reduction in skin thickening (Longevity.Technology, 2026). An early, open-label study in actinic keratoses (rough precancerous skin patches) reported a 46% fall in lesion count at four weeks (Rubedo, 2026). Skin is a sensible place to start. You can see it, biopsy it and dose it locally without exposing the whole body.

Not every "longevity pill" is AI-made#

Retro Biosciences started a Phase 1 trial of RTR242, an autophagy-boosting pill, in Adelaide, Australia, at the end of 2025. Coverage of that trial describes no AI involvement in the drug's discovery; Retro's AI work with OpenAI sits in a separate cell-reprogramming programme (Longevity.Technology, 2025). Readers should check which claim applies to which product. The label "AI longevity company" can cover very different things.

Where AI genuinely helps, and where it does not#

The pipeline in the lab#

Academic groups have shown that machine learning can find senolytics far more cheaply than brute-force screening. A team at MIT screened 2,352 compounds, trained graph neural networks on the results, then scored more than 800,000 molecules. Three candidates selectively killed senescent cells, and one reduced senescent cell burden in the kidneys of aged mice (Wong et al., 2023). Some of those authors are affiliated with Integrated Biosciences, a company built on this approach. In Edinburgh, a separate group trained simple models only on published data and identified ginkgetin, periplocin and oleandrin as senolytics, reporting a several hundred-fold reduction in screening costs (Smer-Barreto et al., 2023).

The pipeline in people#

Clinical success is a harder test. An analysis by Boston Consulting Group researchers found that AI-discovered molecules passed Phase 1 at a rate of 80% to 90%, well above historical averages, but Phase 2 success was around 40%, roughly the industry norm, on a small sample (Jayatunga et al., 2024). That pattern makes sense. AI is good at designing molecules that behave like drugs: absorbed, stable, not obviously toxic. Whether hitting a target actually helps patients is a question about biology, and biology still has the last word.

A 2026 editorial in Expert Opinion on Drug Discovery put it neatly: AI is "unlikely to lower failure rates markedly", but it can make uncertainty measurable and help companies decide which bets to make (Moro, 2026). For longevity, where the biology is especially murky, that may be the most realistic promise.

What would convince a sceptic?#

If I were advising a graduate student or a founder in this space, I would watch for four things over the next few years.

1. Hard outcomes in the disease trial. If GENESIS-IPF-3 slows FVC decline over 52 weeks, rentosertib becomes a credible IPF drug regardless of any ageing story (Insilico Medicine, 2026).
2. Ageing biomarkers in people without the disease. The rentosertib authors themselves call for replication outside IPF (Longevity.Technology, 2026).
3. Clocks that are validated against outcomes. A clock that predicts mortality in a biobank, as ProtAge does, is useful (Argentieri et al., 2024). Showing that a drug-induced change in that clock predicts fewer diseases is a much higher bar, and nobody has cleared it yet.
4. Safety over years, not weeks. Longevity drugs would be taken by largely healthy people for a long time. Liver signals in both rentosertib and azelaprag studies show why that matters (Xu et al., 2025; BioAge, 2024).

The honest summary is mixed. AI drug discovery has delivered molecules that are safe enough to test and, in one case, effective enough to justify a Phase 3 trial. It has not yet delivered a drug proven to slow human ageing. The dual-purpose trial design is a clever way to start collecting that evidence, and I would rather see companies publish clock data with all its caveats than make claims with no data at all.

Frequently asked questions#

Has any AI-designed drug been proven to slow human ageing? No. Rentosertib lowered predicted biological age on six proteomic clocks in a small, 12-week analysis of lung disease patients, but the authors say the clocks cannot separate an anti-ageing effect from the benefit of treating the disease (Zhavoronkov et al., 2026).

What is rentosertib approved for? Nothing yet. It is an investigational drug for idiopathic pulmonary fibrosis, now in a Phase 3 trial of about 320 patients in China (Insilico Medicine, 2026).

What does "AI-designed" actually mean for these drugs? It varies. Insilico used AI both to pick the TNIK target and to generate the molecule (Ren et al., 2024). Other companies use AI or large datasets mainly to choose targets, then use conventional chemistry.

Are ageing clocks reliable enough to judge a drug? They are promising research tools. Some proteomic clocks predict mortality and age-related disease in large biobanks (Argentieri et al., 2024), but no regulator accepts a change in a clock as proof of benefit.

Do AI-discovered drugs succeed more often in clinical trials? They seem to pass Phase 1 safety tests more often (80% to 90%), but Phase 2 success is around 40%, similar to conventional drugs, based on a limited sample (Jayatunga et al., 2024).

What are senolytics, and are any AI-found ones in people? Senolytics clear senescent "zombie" cells. Rubedo's RLS-1496, developed with an AI platform, is in early skin trials with preliminary positive results (Rubedo, 2026).

Why don't longevity companies just run trials on ageing itself? Because regulators approve drugs for specific diseases and lifespan trials would take decades, companies test drugs in age-related diseases and add ageing biomarkers as extra measurements (Zhavoronkov et al., 2026).

Should I try to get these drugs? No. They are experimental, available only in regulated trials, and have known side effects, including liver problems in some participants (Xu et al., 2025). Speak to a doctor about evidence-based ways to protect your health.

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
2. 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. 2025;31(8):2602-2610. https://doi.org/10.1038/s41591-025-03743-2
3. Ren F, Aliper A, Chen J, et al. A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models. Nature Biotechnology. 2024;43(1):63-75. https://doi.org/10.1038/s41587-024-02143-0
4. López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G. Hallmarks of aging: an expanding universe. Cell. 2023;186(2):243-278. https://doi.org/10.1016/j.cell.2022.11.001
5. Argentieri MA, Xiao S, Bennett D, et al. Proteomic aging clock predicts mortality and risk of common age-related diseases in diverse populations. Nature Medicine. 2024;30(9):2450-2460. https://doi.org/10.1038/s41591-024-03164-7
6. Agboola OE, Agboola SS, Adedeji OS, et al. A critical appraisal of structure-based virtual screening, binding free energy methods, and AI-driven molecular design across the evolving computational landscape of drug discovery. Journal of Computer-Aided Molecular Design. 2026;40(1). https://doi.org/10.1007/s10822-026-00955-7
7. Xu J, Liu X, Liang X, Cai X, Qian W. Comparative efficacy and safety of monotherapy and combination pharmacotherapies for idiopathic pulmonary fibrosis: a network meta-analysis of randomized controlled trials. BMC Pulmonary Medicine. 2026;26(1). https://doi.org/10.1186/s12890-026-04342-8
8. Kp Jayatunga M, Ayers M, Bruens L, Jayanth D, Meier C. How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discovery Today. 2024;29(6):104009. https://doi.org/10.1016/j.drudis.2024.104009
9. Wong F, Omori S, Donghia NM, Zheng EJ, Collins JJ. Discovering small-molecule senolytics with deep neural networks. Nature Aging. 2023;3(6):734-750. https://doi.org/10.1038/s43587-023-00415-z
10. Smer-Barreto V, Quintanilla A, Elliott RJR, et al. Discovery of senolytics using machine learning. Nature Communications. 2023;14(1):3445. https://doi.org/10.1038/s41467-023-39120-1
11. Moro S. Incentivizing challenging drug discovery: from reducing failure to quantifying uncertainty. Expert Opinion on Drug Discovery. 2026;21(10):1043-1049 (editorial). https://doi.org/10.1080/17460441.2026.2733199
12. Insilico Medicine. Insilico Medicine doses first patient in GENESIS-IPF-3 Phase III trial (press release, 10 September 2026). https://insilico.com/news/isn1009261-insilico-medicine-doses-first-patient-genesis-ipf-3
13. BioAge Labs. BioAge reports positive Phase 1 data for BGE-102, a novel oral NLRP3 inhibitor (press release, 21 April 2026). https://ir.bioagelabs.com/news-releases/news-release-details/bioage-reports-positive-phase-1-data-bge-102-novel-oral-nlrp3
14. BioAge Labs. Discontinuation of the STRIDES Phase 2 study of azelaprag (press release, 6 December 2024). https://bioagelabs.gcs-web.com/node/7081/pdf
15. Rubedo Life Sciences. First patient dosed with RLS-1496 in Phase 1 clinical trial (press release, 22 May 2025). https://www.businesswire.com/news/home/20250522955270/en/
16. Rubedo Life Sciences. RLS-1496 reduces actinic keratosis lesions by 46% at four weeks in preliminary Phase 1b/2a results (press release, 28 May 2026). https://www.biospace.com/press-releases/rubedos-rls-1496-reduces-actinic-keratosis-pre-cancerous-skin-lesions-by-46-at-four-weeks-with-minimal-irritation-in-preliminary-results-of-phase-1b-2a-study
17. Longevity.Technology. Rentosertib puts aging clocks to the clinical test (science journalism, 8 September 2026). https://longevity.technology/news/rentosertib-puts-aging-clocks-to-the-clinical-test/
18. Longevity.Technology. Rubedo reports positive Phase 1 results for RLS-1496 (science journalism, 27 March 2026). https://longevity.technology/news/rubedo-reports-positive-phase-1-results-for-rls-1496/
19. Longevity.Technology. Retro Bio commences first-in-human trial (science journalism, 30 December 2025). https://longevity.technology/news/retro-bio-commences-first-in-human-trial/
20. Bio-IT World. Insilico Medicine launches Phase III trial for AI-developed drug (8 July 2026). https://www.bio-itworld.com/news/2026/07/08/insilico-medicine-launches-phase-iii-trial-for-ai-developed-drug
21. BioSpace. Roche's Chugai breaks into aging via potential $1B+ deal with AI outfit Gero (7 July 2025). https://www.biospace.com/business/roches-chugai-breaks-into-aging-via-potential-1b-deal-with-ai-outfit-gero