Rapamycin, Metformin or AI? The 2026 Evidence on Longevity Drugs

An immune drug, a diabetes pill and a neural network walk into a clinic#

Ask a room of longevity enthusiasts what they take, and two names come up more than any others. One is rapamycin, a drug that damps down a cell growth switch called mTOR and, in sicker patients, raises the risk of infection (Lee et al., 2024). The other is metformin, the standard first-line tablet for type 2 diabetes (Bannister et al., 2014). In 2026 a third contender has joined them: drugs and drug candidates found with artificial intelligence, sometimes accompanied by AI-built "clocks" that claim to measure how fast a body is ageing.

All three have loud fans. Each has also been oversold at some point. So rather than ask which one is exciting, this piece asks a plainer question: what does the evidence support today, and where does it stop?

The short version is uncomfortable for everyone. Rapamycin has the best animal data of any drug tested against ageing, but its human trials are small and short. Metformin's case has got weaker, not stronger, as researchers have looked harder. And AI has produced real molecules and real clinical results, but nothing yet that shows a person ageing more slowly in any way that matters to their health.

The basics: what "slowing ageing" means to a scientist#

Before comparing drugs, it helps to agree on the target. Geroscience is the field built on the idea that ageing itself is the biggest single risk factor for heart disease, cancer, dementia and frailty, so slowing the underlying biology might delay many diseases at once. That idea is the founding logic of the TAME trial, which proposed using metformin as a first test of whether a drug could delay several age-related diseases together (Barzilai et al., 2016).

A few terms come up again and again.

• Lifespan is how long an organism lives. Healthspan is how long it lives in good health. Most researchers care more about the second.
• mTOR (mechanistic target of rapamycin) is a protein that acts like a cell's growth switch. When nutrients are plentiful it tells cells to build and divide; when it is turned down, cells shift towards maintenance and recycling. Rapamycin dials mTOR down.
• Autophagy is that recycling process, where cells break down worn-out parts.
• Senescent cells are damaged cells that stop dividing but do not die, and which release inflammatory signals. Drugs that kill them are called senolytics.
• A biological age clock is a statistical model, often built with machine learning, that estimates someone's "biological age" from blood proteins, DNA chemical tags (methylation) or other measurements. If your clock reads 55 and you are 50, the model thinks your body looks older than average for your age.
• A randomised controlled trial (RCT) assigns people by chance to a drug or a placebo, which is the most reliable way to tell whether the drug caused any difference. Observational studies simply compare people who happened to take a drug with people who did not, and they are prone to hidden biases.

Keep those last two in mind. A lot of the longevity story turns on the gap between them.

Rapamycin longevity: the strongest animal case in the field#

The modern rapamycin story starts with one mouse study. In 2009 the US National Institute on Aging's Interventions Testing Program (ITP), which tests compounds at three separate laboratories in genetically mixed mice, reported that rapamycin extended both median and maximum lifespan even when feeding began at 600 days of age, roughly the equivalent of a 60-year-old human (Harrison et al., 2009). Measured at the age when 90% of the animals had died, lifespan rose by 14% in females and 9% in males (Harrison et al., 2009).

That result held up. A later ITP study using a threefold higher dose raised median lifespan by 23% in males and 26% in females, and found that females benefited more at every dose tested (Miller et al., 2014). The same paper found that rapamycin's metabolic effects differed from those of dietary restriction, so it is not simply "fasting in a pill" (Miller et al., 2014).

The broadest test so far came in 2025. A meta-analysis pooling 911 effect sizes from 167 papers across eight vertebrate species found that rapamycin, but not metformin, produced significant lifespan extension, comparable to that of dietary restriction (Ivimey-Cook et al., 2025). The authors also flagged high heterogeneity and significant publication bias across all treatments, which means the true effect sizes may be smaller than the published average (Ivimey-Cook et al., 2025).

The chart below shows the headline ITP mouse figures side by side with metformin.

Figure: Mouse lifespan gains reported by the NIA Interventions Testing Program. Data from Harrison et al., 2009, Miller et al., 2014 and Strong et al., 2016.

What happens when people take it?#

Here the evidence thins out quickly. The earliest encouraging human data came from Novartis, where a rapamycin-like drug called everolimus improved the response to flu vaccination by about 20% in older volunteers at well-tolerated doses (Mannick et al., 2014). A follow-up trial in 264 older adults found that six weeks of low-dose mTOR inhibition was linked to fewer reported infections over the following year (Mannick et al., 2018).

The most direct test in healthy people is PEARL, a 48-week, double-blind, placebo-controlled trial of 5 mg or 10 mg of compounded rapamycin taken once a week (Moel et al., 2025). Its primary outcome, visceral fat, did not change (Moel et al., 2025). Women on 10 mg gained lean tissue and reported less pain, and the 5 mg group reported better emotional wellbeing and general health, while adverse events were similar across all groups (Moel et al., 2025). Those are secondary findings in subgroups, which is exactly the kind of result that often fails to replicate. The authors themselves describe the regimen as relatively safe and say future work must establish efficacy more comprehensively (Moel et al., 2025).

Rapamycin side effects: the honest ledger#

A 2024 systematic review in The Lancet Healthy Longevity pulled together 19 human studies of rapamycin and its relatives (Lee et al., 2024). It found improvements in markers of ageing in the immune, cardiovascular and skin systems, no significant effect on endocrine, muscle or brain measures, and no serious adverse events attributed to the drugs in healthy people. In people with age-related diseases, though, the drugs were linked to more infections and higher cholesterol and triglycerides. The review also noted that long-term effects remain largely untested (Lee et al., 2024).

So people searching for "rapamycin for longevity" or the right "rapamycin dose for longevity" are, in effect, running an uncontrolled experiment on themselves. Nobody yet knows the best dose, schedule or age to start, and no trial has shown that rapamycin prevents disease or extends life in humans.

Metformin longevity: a promising story that keeps getting smaller#

Metformin's anti-ageing reputation rests largely on one striking observational result. A 2014 analysis of UK primary care records found that people with type 2 diabetes who started on metformin had a longer median survival than matched people without diabetes (Bannister et al., 2014). If a diabetes drug could make diabetics outlive healthy people, the thinking went, it might do something special for everyone.

That claim has not aged well. A Danish registry study, including comparisons between twins where only one had diabetes, found the reverse: starting metformin for type 2 diabetes was associated with higher mortality than having no diabetes, even after adjustment for other illnesses and social factors (Keys et al., 2022). The authors concluded that metformin users did not survive as long as, or longer than, the general population (Keys et al., 2022).

The mouse data are no kinder. In the same ITP programme that made rapamycin famous, metformin alone at 0.1% of the diet did not significantly extend lifespan (Strong et al., 2016). Interestingly, metformin combined with rapamycin did extend lifespan robustly, though that judgement relied on comparison with earlier rapamycin-only studies rather than a direct head-to-head (Strong et al., 2016). And the 2025 vertebrate meta-analysis found no significant lifespan extension from metformin across species (Ivimey-Cook et al., 2025).

There is also a practical catch for the gym-going crowd. In a randomised trial of older adults doing 12 weeks of aerobic training, metformin blunted the improvements in insulin sensitivity and aerobic fitness (VO2max) that exercise normally brings, and cancelled out the boost in muscle mitochondrial respiration (Konopka et al., 2019). For a healthy person, swapping some of exercise's benefit for an unproven drug effect is a questionable trade.

The case for metformin is not dead#

It would be unfair to stop there. A 2024 study in Cell gave metformin to adult male cynomolgus monkeys for 40 months and reported slower ageing across several molecular clocks, including a roughly six-year regression in brain ageing measures (Yang et al., 2024). That is a primate result, which counts for a lot, though it involved only males and relied on clocks the team built themselves (Yang et al., 2024).

Human data also offer some support when the comparison group is chosen carefully. Among women with type 2 diabetes in the Women's Health Initiative, starting metformin rather than a sulfonylurea was linked to a 30% lower risk of dying before age 90 (Shadyab et al., 2025). The authors were explicit that the observational design leaves room for confounding, so causality cannot be inferred (Shadyab et al., 2025). Note too that this compares two diabetes drugs; it says nothing about giving metformin to people without diabetes.

What happened to TAME?#

The trial meant to settle the question still has not happened. In August 2025, lead investigator Nir Barzilai told Science|Business that pharmaceutical companies showed no interest because metformin cannot be patented, and that a private foundation had agreed to fund only a third of the trial (Science|Business, 2025). A September 2026 report described TAME as an ageing trial that had never enrolled a patient (Technology.org, 2026). For a drug that costs pennies a day, the absence of a definitive trial after a decade is its own kind of evidence about how medical research is financed.

AI enters the race: new molecules, new measuring sticks#

AI is doing two separate jobs in longevity research, and it helps to keep them apart. One is finding drug candidates. The other is building tools, like ageing clocks, to judge whether any drug works.

Finding molecules#

The clearest example of AI-led discovery reaching patients is rentosertib, a drug from Insilico Medicine aimed at an enzyme called TNIK, with both the target and the molecule identified using generative AI. In a 12-week phase 2a trial of 71 people with idiopathic pulmonary fibrosis, a scarring lung disease strongly linked to age, the 60 mg once-daily group gained an average of 98.4 mL of forced vital capacity (a standard measure of how much air the lungs can push out), while the placebo group lost 20.3 mL (Xu et al., 2025). Liver toxicity and diarrhoea were the most common reasons people stopped treatment (Xu et al., 2025). That is a genuine clinical signal, in a small trial, for a lung disease. It is not an anti-ageing result.

On the discovery side, an MIT team screened 2,352 compounds for senolytic activity, trained graph neural networks on the results, and then used them to predict activity across more than 800,000 molecules (Wong et al., 2023). One of the hits reduced the burden of senescent cells in the kidneys of aged mice (Wong et al., 2023). That is a neat demonstration that AI can widen the search, but it remains a mouse result.

Meanwhile the start-up Retro Biosciences began a phase 1 safety trial in healthy volunteers in Adelaide, Australia, of RTR242, a pill designed to restore lysosomal function, with Alzheimer's disease named as a possible first indication (Longevity.Technology, 2025). Phase 1 trials test safety, not whether a drug works.

Measuring ageing with AI clocks#

The second AI job is measurement, and this is where hype and science collide most often. In September 2026, Nature Biotechnology published an analysis of blood samples from 42 participants in the rentosertib lung trial, scored with six machine-learning proteomic ageing clocks (Zhavoronkov et al., 2026). Press coverage reported predicted age reductions of roughly three to four years at week four in some dose groups (News-Medical, 2026).

Read the caveats before the headline. The participants were patients with lung fibrosis, not healthy volunteers. A fibrosis-related protein called LTBP2 contributed heavily to all six clocks, so their agreement may mostly reflect the drug's anti-scarring effect rather than slower ageing (Lifespan.io, 2026). And the company that makes the drug also developed one of the clocks (Lifespan.io, 2026). A change in a biological age score is a change in a model's output. Whether it predicts longer, healthier life in treated people is a separate question that nobody has yet answered for any drug.

Where the evidence stands: a side-by-side scorecard#

The bottom row is the one that matters most, and it reads the same across all three columns.

What comes next#

Two larger efforts may change the picture. The Dog Aging Project's TRIAD study is giving low-dose rapamycin to older companion dogs over three years, with 170 dogs enrolled by late 2024 and a target of 580 (Texas A&M, 2024). Dogs share our homes and many of our environmental exposures, which makes them a useful halfway house between mice and people.

In humans, ARPA-H is funding VITAL-H, a trial of up to $38 million led by UT Health San Antonio that will test rapamycin, dapagliflozin and semaglutide against placebo in 726 adults in their sixties (UT Health San Antonio, 2026). Its main aim is to validate "intrinsic capacity", a composite measure of physical and mental function, as an outcome for ageing trials (UT Health San Antonio, 2026). That may prove as valuable as any single drug result, because the field badly needs agreed ways to measure success that regulators will accept.

My own read is this. Rapamycin is the most credible candidate, and the honest description of its human evidence is "promising and unfinished". Metformin is a reasonable drug for diabetes whose anti-ageing promise has shrunk under scrutiny, and it deserves the trial it never got. AI is speeding up discovery and giving us sharper instruments, but a clock that ticks backwards in a blood test is a hypothesis, not a result. In 2026, the best-supported longevity interventions are still the dull ones: exercise, not smoking, sleep, and treating blood pressure and cholesterol.

Frequently asked questions#

Is rapamycin safe to take for longevity? In healthy adults, short trials have not found serious adverse events attributed to rapamycin, but people with existing illnesses have shown more infections and higher blood lipids, and long-term effects remain untested (Lee et al., 2024). It is a prescription immunosuppressant and should only be used under medical supervision.

What dose of rapamycin is used in longevity trials? The PEARL trial tested 5 mg and 10 mg taken once a week for 48 weeks (Moel et al., 2025). No trial has established an optimal dose for slowing ageing.

Does metformin slow ageing in people without diabetes? There is no randomised trial showing this. Metformin alone did not significantly extend mouse lifespan (Strong et al., 2016), and a Danish registry study found metformin users with diabetes did not outlive people without diabetes (Keys et al., 2022).

Can I take metformin and still get the benefits of exercise? In older adults doing aerobic training, metformin reduced the usual gains in insulin sensitivity and fitness (Konopka et al., 2019). Anyone considering it should discuss this trade-off with a doctor.

Has an AI-designed drug been shown to reverse biological age? An analysis of the rentosertib lung fibrosis trial reported lower scores on proteomic ageing clocks (Zhavoronkov et al., 2026), but the patients had lung disease, the sample was 42 people, and the clocks may be tracking the drug's anti-fibrosis effect (Lifespan.io, 2026). It is not proof of slower ageing.

Are biological age tests reliable enough to judge a longevity drug? Not yet on their own. Clocks can shift for reasons unrelated to ageing, and the field is still working to validate outcomes such as intrinsic capacity that regulators would accept (UT Health San Antonio, 2026).

Why has the TAME metformin trial not started? Its investigators report that drug companies have little incentive to fund a trial of an unpatentable generic, and only part of the money has been secured (Science|Business, 2025).

Which longevity drug has the strongest evidence in 2026? Rapamycin has the most consistent animal evidence, including a cross-species meta-analysis (Ivimey-Cook et al., 2025). No drug, including rapamycin, has yet been shown to extend healthy human life.

This article is for information only and is not medical advice. Do not start, stop or change any medication without speaking to a qualified clinician.

References#

1. Barzilai N, Crandall JP, Kritchevsky SB, Espeland MA. Metformin as a tool to target aging. Cell Metabolism. 2016;23(6):1060-1065. https://doi.org/10.1016/j.cmet.2016.05.011
2. Harrison DE, Strong R, Sharp ZD, et al. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. Nature. 2009;460:392-395. https://doi.org/10.1038/nature08221
3. Miller RA, Harrison DE, Astle CM, et al. Rapamycin-mediated lifespan increase in mice is dose and sex dependent and metabolically distinct from dietary restriction. Aging Cell. 2014;13(3):468-477. https://doi.org/10.1111/acel.12194
4. Strong R, Miller RA, Antebi A, et al. Longer lifespan in male mice treated with a weakly estrogenic agonist, an antioxidant, an α-glucosidase inhibitor or a Nrf2-inducer. Aging Cell. 2016;15(5):872-884. https://doi.org/10.1111/acel.12496
5. Ivimey-Cook ER, Sultanova Z, Maklakov AA. Rapamycin, not metformin, mirrors dietary restriction-driven lifespan extension in vertebrates: a meta-analysis. Aging Cell. 2025;24(9):e70131. https://doi.org/10.1111/acel.70131
6. Mannick JB, Del Giudice G, Lattanzi M, et al. mTOR inhibition improves immune function in the elderly. Science Translational Medicine. 2014;6(268):268ra179. https://doi.org/10.1126/scitranslmed.3009892
7. Mannick JB, Morris M, Hockey HP, et al. TORC1 inhibition enhances immune function and reduces infections in the elderly. Science Translational Medicine. 2018;10(449):eaaq1564. https://doi.org/10.1126/scitranslmed.aaq1564
8. Moel M, Harinath G, Lee V, et al. Influence of rapamycin on safety and healthspan metrics after one year: PEARL trial results. Aging (Albany NY). 2025;17(4):908-936. https://www.aging-us.com/article/206235
9. Lee DJW, Hodzic Kuerec A, Maier AB. Targeting ageing with rapamycin and its derivatives in humans: a systematic review. The Lancet Healthy Longevity. 2024;5(2):e152-e162. https://tinyurl.com/3dh2zxm5
10. Bannister CA, Holden SE, Jenkins-Jones S, et al. Can people with type 2 diabetes live longer than those without? Diabetes, Obesity and Metabolism. 2014;16(11):1165-1173. https://doi.org/10.1111/dom.12354
11. Keys MT, Thinggaard M, Larsen LA, et al. Reassessing the evidence of a survival advantage in type 2 diabetes treated with metformin compared with controls without diabetes: a retrospective cohort study. International Journal of Epidemiology. 2022;51(6):1886-1898. https://doi.org/10.1093/ije/dyac200
12. Konopka AR, Laurin JL, Schoenberg HM, et al. Metformin inhibits mitochondrial adaptations to aerobic exercise training in older adults. Aging Cell. 2019;18(1):e12880. https://doi.org/10.1111/acel.12880
13. Yang Y, Lu X, Liu N, et al. Metformin decelerates aging clock in male monkeys. Cell. 2024;187(22):6358-6378. https://doi.org/10.1016/j.cell.2024.08.021
14. Shadyab AH, Espeland MA, Odegaard AO, et al. Comparative effectiveness of metformin versus sulfonylureas on exceptional longevity in women with type 2 diabetes: target trial emulation. Journals of Gerontology Series A. 2025;80(7). https://doi.org/10.1093/gerona/glaf095
15. 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. https://doi.org/10.1038/s41591-025-03743-2
16. Zhavoronkov A, 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
17. 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
18. Texas A&M School of Veterinary Medicine. Dog Aging Project TRIAD press release. 9 December 2024. https://vetmed.tamu.edu/news/press-releases/dog-aging-project-triad/
19. UT Health San Antonio. UT San Antonio to lead $38 million national trial testing drugs to extend healthspan. 4 March 2026. https://news.uthscsa.edu/ut-san-antonio-to-lead-38-million-national-trial-testing-drugs-to-extend-healthspan/
20. Science|Business. Anti-ageing therapies: have we got health research funding wrong? 7 August 2025 (science journalism; primary funding records not publicly available). https://sciencebusiness.net/news/drug-development/anti-ageing-therapies-have-we-got-health-research-funding-wrong
21. Technology.org. TAME: the aging trial that never enrolled a patient. 26 September 2026 (science journalism). https://www.technology.org/2026/09/26/tame-trial-metformin-aging-funding/
22. Longevity.Technology. Retro Bio commences first-in-human trial. 30 December 2025 (science journalism; no registry entry located). https://longevity.technology/news/retro-bio-commences-first-in-human-trial/
23. News-Medical. AI-designed drug candidate reverses biological age in clinical study. 7 September 2026 (sponsor press release). https://www.news-medical.net/news/20260907/AI-designed-drug-candidate-reverses-biological-age-in-clinical-study.aspx
24. Lifespan.io. Lung fibrosis drug lowers biological age estimates. 2026 (science journalism summarising reference 16). https://lifespan.io/?p=158777