Jeff Bezos and Yuri Milner backed Altos Labs with $3 billion in 2022. Sam Altman put $180 million into Retro Biosciences the same year. Peter Thiel has been backing longevity research through the Methuselah Foundation and related vehicles since the early 2000s. The convergence of enormous private wealth and aging biology — the nascent science that aging itself is a treatable disease with definable molecular mechanisms — has created an unusual sector where the ambition is not to treat any specific disease but to slow or reverse the biological process that underlies all of them.
AI has entered this space as an accelerant. The stated thesis: the biology of aging involves complex interactions across thousands of molecular pathways that are too intricate for human scientists to navigate intuitively, but that ML models might navigate by identifying patterns in large multi-omics datasets. Combine this computational horsepower with reprogramming technologies (partial cellular reprogramming based on Shinya Yamanaka’s Nobel Prize-winning work), senolytics (drugs that clear senescent cells), and other candidate mechanisms, and you have the infrastructure for a new kind of pharmaceutical program.
The thesis is not crazy. What it is is genuinely difficult to evaluate against the actual biology.
The Hallmarks Framework
The dominant conceptual framework for the biology of aging comes from a 2013 paper by López-Otín et al. in Cell, which described nine “hallmarks of aging” — categories of molecular and cellular dysfunction that accumulate with age and collectively drive the phenotypes of aging: genomic instability, telomere attrition, epigenetic alterations, loss of proteostasis, deregulated nutrient sensing, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, and altered intercellular communication. An updated version published in 2023 added three more hallmarks, reflecting expanding understanding.
This framework is widely taught and widely used. It’s also a categorization, not a mechanism. Saying that “epigenetic alterations” drive aging is like saying “communication breakdown” drives organizational failure — it’s true and it’s incomplete. The specific epigenetic changes, their tissue-specific dynamics, their causal relationships to functional decline, and the question of which ones are causes versus consequences of aging — these remain active research questions that the hallmarks framework helpfully organizes but does not answer.
Related reading
AI applications in longevity research have focused heavily on the “epigenetic clock” work pioneered by Steve Horvath at UCLA. Methylation patterns on DNA change with age in systematic ways that can be measured and that correlate with biological rather than chronological age. Several AI models have been built to measure “biological age” from DNA methylation data, blood proteomics, or clinical measurements — and some of these “aging clocks” have been validated as predictors of age-related disease risk.
The leap from measuring biological age accurately to identifying interventions that durably slow its progression is where the evidence thins considerably.
The Senolytic Story
The best evidence for a pharmacological intervention targeting an aging mechanism comes from senolytics — drugs that selectively eliminate senescent cells (cells that have stopped dividing but remain metabolically active and secrete inflammatory molecules into surrounding tissue). The dasatinib plus quercetin combination, first published by the Mayo Clinic’s James Kirkland in 2015, showed that clearing senescent cells in mice extended lifespan and improved several age-related phenotypes.
Human clinical trials of senolytics are underway at Mayo, the Unity Biotechnology clinical program, and several academic sites. Unity Biotechnology’s UBX0101 (a first-generation MDM2 inhibitor senolytic targeting senescent cells in knee osteoarthritis) failed Phase II in 2020 — a significant setback that led to reformulation and a more targeted approach. Unity’s second-generation programs, targeting senescent cells in the retina (UBX1325 for diabetic macular edema and age-related macular degeneration), have shown more promising Phase I/IIa data as of 2025, with durability signals that are being tracked in ongoing studies.
The senolytic story is instructive for the longevity field broadly: the mechanism is coherent, the animal data is strong, the first generation of clinical translation encountered the same attrition that first-generation translations encounter in nearly every therapeutic area, and the second generation is producing more refined and promising signals. This is normal drug development at a multi-year timescale. It does not require AI to proceed, though AI-assisted patient selection and biomarker development is being applied to ongoing programs.
Where AI Is Contributing
The clearest contribution of AI to longevity research is in the identification and validation of molecular targets — finding which genes, proteins, or pathways, when modulated, produce the most robust lifespan or healthspan benefits across model organisms and translatable to human biology.
Unity, Altos Labs, Calico (Google’s longevity subsidiary), and the Buck Institute for Research on Aging are all generating large multi-omics datasets from aging model organisms and human samples. Training ML models on these datasets to identify targets that appear consistently across species is a legitimate use of AI’s pattern-recognition capabilities. The targets identified this way are starting points for biology, not drugs — there’s still a full drug development process required — but the starting points may be better because the ML models can integrate more complex prior data than a human research team reviewing the literature.
Insilico Medicine, better known for its IPF program, has also published AI-identified aging targets in less-studied pathways. Several of these targets have academic validation studies underway. How many will survive the transition to druggable lead compounds remains unclear.
The Regulatory Void
Here is the most consequential structural problem in longevity drug development: the FDA does not recognize aging as an indication. You cannot run a clinical trial with “delay of aging” or “extension of healthspan” as your primary endpoint, because there’s no agreed regulatory definition of successful treatment. Every longevity drug program must, therefore, use a specific disease endpoint — cardiovascular disease, osteoporosis, kidney disease, dementia — as its regulatory pathway, even if the underlying hypothesis is about aging biology.
Related reading
This is not merely bureaucratic obstruction. It reflects a genuine scientific problem: how do you demonstrate, within a 3-5 year clinical trial, that a drug has slowed the process of aging? The FDA’s Targeting Aging with Metformin (TAME) trial — studying whether the generic diabetes drug metformin delays age-related diseases in healthy older adults — is the first attempt at this with a specific protocol design acceptable to the FDA. It began enrolling in 2023 and won’t have primary outcome data until roughly 2028-2030. That timeline is the honest measure of how long it takes to demonstrate aging-relevant outcomes even when the regulatory framework exists.
The Demographic Context
The longevity sector’s most direct commercial competition is not other drug companies — it’s life insurance actuaries. The insurance industry has priced human longevity risk for 200 years with a fair degree of accuracy, which means that durable improvements in human lifespan (as opposed to specific disease treatment) would create large and complex economic dislocations in pension systems, healthcare financing, and social security structures that were designed around historical mortality curves.
This context is rarely discussed in the longevity investment thesis, which focuses on the technology. But the social and economic adaptation required if longevity drugs work as their proponents hope would be at least as challenging as the biological problem they solve. That’s not a reason to abandon the research. It is a reason to be honest about what “success” in longevity drug development actually involves beyond the science.
AI will contribute to the biology. The biology, if it succeeds, will create problems that no algorithm can address.
The Data Flywheel Opportunity
One place where AI and longevity research have a genuinely productive intersection is longitudinal data analysis. Understanding what predicts healthy aging — as opposed to merely predicting longevity — requires following large populations across decades, measuring thousands of biological and behavioral variables, and identifying the patterns that discriminate healthy agers from those who develop age-related disease early.
The UK Biobank (500,000 participants, now with longitudinal follow-up data extending 15-20 years), the Human Longevity Institute’s database, and the All of Us Research Program in the United States are generating datasets of unprecedented breadth. ML models applied to these datasets can identify aging trajectories, discover biomarker combinations that predict healthy aging years before phenotypic changes appear, and generate hypotheses about modifiable risk factors that conventional epidemiology would miss due to the high dimensionality of the data.
This application is conservative relative to the “design drugs that reverse aging” narrative but more likely to generate near-term clinical value. Understanding who is aging well and why — and identifying the biological signatures that distinguish them — is a tractable problem with a good evidence base. The findings can inform drug target selection, clinical trial population design, and eventually public health interventions that don’t require any new drugs at all.
Calico and the Long Game
Calico, Google’s longevity subsidiary operating since 2013 with an undisclosed budget that observers estimate in the hundreds of millions annually, has operated with unusual opacity for a life sciences company. Their published research through 2025 covers aging biology in model organisms, molecular mechanisms of longevity pathways, and computational analysis of aging datasets. They have no clinical-stage programs publicly disclosed.
This opacity has generated industry skepticism about whether Calico is producing translatable science or funding basic research indefinitely without commercial output. The honest interpretation may be more charitable: longevity drug development has a minimum viable timeline that no amount of money or AI can compress, because the evidence required to demonstrate that a drug slows aging in humans requires years of human observation. Calico may simply be on the correct timeline for the biology they’re doing, which happens to look slow relative to the investment cycle that funded them.
Related reading
The longevity field overall is at an early stage relative to the ambition of its claims. The money is real, the biology is genuinely interesting, and some of the science is of high quality. The gap between “interesting biology” and “approved drug that durably extends human healthspan” is at least a decade of clinical work, probably two. AI can shorten some of those timelines at the margins. The biology sets the minimum pace, and it is slower than the fundraising.
The Near-Term Realistic Wins
A more useful frame than “will AI cure aging” is “what specific things can AI contribute to longevity science in the next five years that would be genuinely valuable?” Several answers emerge from the current state of the field.
AI analysis of longitudinal biological aging data — the UK Biobank cohort, the NIH’s All of Us program, the diverse emerging cohorts from Africa and Asia that are underrepresented in current datasets — can identify aging biomarkers with broader population validity than current clocks derived from predominantly white, European study participants. This matters because biological aging rates differ meaningfully across ancestries and environments, and aging clocks calibrated on narrow populations may misclassify aging rates in populations that weren’t included in their training.
AI can improve the design of the TAME trial and its successors by helping identify the patient subpopulations most likely to respond to metformin-based slowing of biological aging, potentially finding a precision medicine signal in a population-level intervention. It can accelerate the identification of synthetic aging clocks — combinations of cheap clinical measurements (blood counts, metabolic panels, blood pressure) that approximate expensive methylation clock measurements, making biological age assessment accessible outside research settings. And it can organize the increasingly complex landscape of longevity intervention trials — which are proliferating in academic medicine — to reduce duplication and identify the most promising mechanistic hypotheses to test in human studies.
None of these wins look like “AI discovers drug that extends lifespan by 20 years.” They look like better science, faster, with broader applicability. That is what AI does for medicine at its best — it accelerates the good science, not the hype.
One email a month: new articles, reviews and the upcoming live webinar + free recording. No spam, unsubscribe anytime.