Altos Labs: $3 Billion, Zero Phase 1 Trials
Clock Science, Mouse Livers
The Horvath DNA methylation clock (Genome Biology, 2013) estimates biological age from methylation patterns at 353 CpG sites (positions in the genome where methyl groups attach, readable from a blood draw) in a tissue sample. The DunedinPACE clock (eLife, 2022) refines the instrument to measure pace of aging rather than a static age estimate: a rate, not a point. Both are correlation instruments calibrated against population datasets. Neither carries a validated clinical endpoint linking slower methylation pace to extended disease-free survival in a randomized human trial. A patient paying for a consumer epigenetic age test receives a prediction, not a clinical finding.
The Israeli partial-reprogramming work reported this week describes rejuvenation of mouse liver cells using Yamanaka-factor combinations (OSK: Oct4, Sox2, and Klf4, transcription factors that transiently reset a cell's epigenetic state toward an earlier configuration). Mouse hepatocytes and human hepatocytes respond differently to factor re-expression; partial reprogramming of human solid-organ tissue carries uncharacterized oncogenic risk. No IND (Investigational New Drug application, the US filing required before human-subject trials) for partial liver reprogramming in humans has cleared regulatory review. Altos Labs raised $3 billion in 2022 specifically for reprogramming biology. Zero Phase 1 trials of human liver reprogramming appear on ClinicalTrials.gov. Altos Labs' public pipeline lists no Phase 1 start date for any human tissue program.
Foundation Models, No Filing
EurekAlert reported this week on an AI foundation model targeting human longevity science. Foundation models trained on multi-omic datasets (proteomics, transcriptomics, methylation profiles) have produced real scientific insight: AlphaFold2 (DeepMind, 2021) compressed years of protein-structure crystallography into days. That capacity is genuine. The clinical decision-support use case requires a different chain: a defined population, a named primary endpoint, IRB (Institutional Review Board, the ethics committee that must approve human-subject research before it begins) approval, and prospective validation in a human cohort. No AI longevity foundation model has publicly completed that chain.
NMPA's Center for Drug Evaluation (China's review division for new drug and device applications) has no published pathway for AI aging-clock output as a clinical claim. TGA's Advisory Committee on Medicines (Australia's pre-registration evaluation committee) has SaMD (Software as a Medical Device) guidance but nothing specific to biological-age prediction output. HSA Singapore classifies AI diagnostic tools under its medical device framework but has not addressed aging-score estimates as a regulated output class. A patient in Singapore following a clinician-prescribed protocol built on an AI aging-clock readout relies on a tool no APAC regulator has reviewed. HSA Singapore has not addressed AI aging-score output as a regulated class. The clinician-prescribed protocol drawing on that output is active.
Diabetes research spent thirty years mapping the biomarker-to-outcome gap. The ACCORD trial (NEJM, 2008, 10,251 patients with type 2 diabetes) found that driving HbA1c below 6% raised mortality rather than reduced it. NMPA, HSA Singapore, TGA, and PMDA have published no regulatory pathway for AI aging-clock output as a clinical claim. The readout is in use in clinician-prescribed protocols in Singapore today. No sponsor has filed the dossier that would require one.