In mice, and only in mice so far. Female mice of one strain treated daily from 20 months of age reached a median lifespan of 834 days against 742 on vehicle, which is 92 days of difference by subtraction from the two published figures [2]. In people, the strongest claim is that three validated DNA methylation clocks decelerated over 32 weeks of semaglutide, and a clock is a model trained to predict aging outcomes rather than a record of them [1]. No trial with a prespecified aging endpoint has been conducted, and the review making the case says so itself.
The longevity market is the next place these drugs are going, and it moves faster than evidence does — faster than the trial evidence that exists can support. This review is a careful version of the argument, which makes it a good place to see exactly where the argument currently stops.
The case, as its authors put it
These drugs reduce major cardiovascular events, all-cause mortality and systemic inflammation in randomized trials [1]. The review cites SELECT’s 19% reduction in all-cause mortality, hazard ratio 0.81, in people with obesity and without diabetes, and FLOW’s 24% reduction in its primary kidney composite, hazard ratio 0.76.
It then makes an inferential move worth marking: it describes those effect sizes as exceeding what glycemic and weight-related improvements alone would predict. That is a comparison against a model of how much benefit weight loss ought to produce, not a measurement of anything. It may well be right, and it is an argument rather than a result.
The clocks
The newest piece is the interesting one. The review reports that randomized evidence published in 2025 showed significant deceleration of three validated DNA methylation clocks — DunedinPACE, PCGrimAge and PhenoAge — over 32 weeks of semaglutide therapy. That finding is described here as the review describes it, and this desk has not read the underlying trial.
The one place lifespan itself was measured
Female mice of a single common laboratory strain were given daily semaglutide by injection from twenty months of age, which is late middle age for a mouse, and treatment continued for the rest of their lives [2]. Median lifespan came out at 742 days on vehicle and 834 days on semaglutide.
What makes that more than an appetite-suppression result is the comparison arm. Eating less is the oldest reliable way to extend life in a laboratory animal, so the authors measured how much less their mice ate, found 24%, and ran a separate group restricted by exactly that 24%. Against it, semaglutide tracked calorie restriction on most measures and did better on exploratory drive, spatial memory and glucose control. The strain, sex and schedule cautions are in the mouse lifespan study.
What the human mortality evidence looks like up close
A target trial emulation in adults with type 2 diabetes and a BMI below 27 found all-cause mortality 39% lower against DPP-4 inhibitors, hazard ratio 0.61 (95% CI 0.52 to 0.71) [3]. Against SGLT2 inhibitors it did not differ at all: 0.89, 95% CI 0.76 to 1.05. The authors describe the first figure as likely residual confounding rather than a benefit of that size.
That pattern is the tell, and it is what a longevity claim built on observational mortality has to get past. When a result is huge against a weak comparator and gone against a strong one, the comparator is doing the work. The reasoning is set out in the hazard ratio its own authors disowned.
The other way the question gets answered is by modeling. A risk equation fitted in 610,789 people was applied to 200,012 survey participants twice, once as observed and once with five risk factors shifted by the amounts SELECT measured [4]. Projected ten-year cardiovascular incidence fell from 13.82% to 10.83%, an absolute reduction of 2.99 percentage points. Nobody in that analysis took any drug, which is set out in the emulated general population.
What the review itself says is missing
Its own list is long and specific: definitive trials with prespecified epigenetic aging endpoints, durability follow-up, body-composition assessment, prespecified sex-stratified analyses, and adequate representation of diverse populations. None of those exist yet.
It flags the pending EVOKE and EVOKE+ readouts in early Alzheimer’s disease as particularly consequential, which is fair, because a hard clinical endpoint in a neurodegenerative disease would say far more than any clock. That question has its own page here.
Why this matters commercially before it matters clinically
Because “slows biological aging” is a claim that can be made about a biomarker long before it can be made about a person, and there is an entire consumer longevity industry built on exactly that gap. Methylation clock tests are already sold direct to consumers, and the pairing of those two products is predictable.
No seller on this roster offers a longevity indication and none claims one. If that changes, the question to ask is the one this review answers honestly about itself: which endpoint moved, and was it a thing that happened to people or a number that predicts things that happen to people? The mechanisms the review names are plausible and preclinical, and mechanistic plausibility is how every failed intervention in this field started. What sellers say now is the baseline, recorded in the disclosure census.