Longevity, genes,
and disease-modifying AI.

ALDH3B2 loss converts human pancreatic duct cells into beta-like cells (Science Translational Medicine; WIRED 2026-09-27)

A genome-wide CRISPR screen found that silencing ALDH3B2 raises duct-to-beta-like conversion to about 8.5% (vs under 1%). Transplanted cells secreted glucose-responsive human insulin and lowered blood glucose toward normal for six weeks in diabetic mice.

Caveat: Preclinical transplant model only; insulin output is below native beta cells; ALDH3B2 is widely expressed, so therapy would need cell-selective delivery. Not a cure claim.

REGENXBIO sura-vec: 3-year durability in NPDR (Phase II ALTITUDE; Retina Society 2026-09-24)

A one-time, in-office suprachoroidal AAV8 anti-VEGF gene therapy (surabgene lomparvovec / ABBV-RGX-314). At Dose Level 3, 60% of participants with 3-year visits (6/10) achieved a 2-step or greater DRSS improvement without additional DR treatment and with no vision-threatening events. No new related safety signals, and no intraocular inflammation with short-course topical steroid prophylaxis. DL3 is now in the Phase IIb/III NAAVIGATE trial.

Caveat: Small long-term subset; investigational; pivotal trial ongoing; not an approval.

YolTech YOLT-202: first-in-human adenine base editing proof of concept in PiZZ AATD (ERS late-breaker, 2026-09-07)

Open-label investigator-initiated study (n=4) of a single-dose LNP adenine base editor that corrects PiZ to PiM. At 45 mg, mean total AAT rose from about 5 to about 20 µM (above the 11 µM protective threshold), durable through 6 months. Corrected M-AAT reached about 93–99% of circulating AAT, and liver biopsy showed up to 57% on-target editing with no reported bystander or off-target edits. Only transient ALT/AST rises.

Caveat: Tiny open-label study; not an FDA approval; different sponsor and platform from Prime Medicine PM647 (yesterday’s lead, which was an IND clearance only).

Nature news: two child deaths in Chinese gene-editing trials and where the field goes next

Reporting and analysis of two pediatric deaths linked to high-dose AAV-delivered gene-editing therapies in investigator-initiated trials in China, and the regulatory push (including tighter oversight of investigator-initiated trials) for faster serious-adverse-event disclosure, dosing discipline, and closer scrutiny of delivery vectors.

Caveat: News analysis for clinician risk framing, not a product approval.

Longeveron + Miller School reach the XPRIZE Healthspan finals with the laromestrocel aging-frailty program (2026-09-14)

A $1M milestone award to test the investigational mesenchymal stem cell (MSC) therapy laromestrocel for aging frailty (muscle, cognition, and immune function). Randomized trial enrollment is expected in summer 2027, building on earlier 6-minute-walk signals.

Caveat: A competition milestone and planned trial, not an efficacy readout or approval; not a lifespan claim.

Research AI ≠ clinical cure claims

A preclinical mouse transplant study, a small long-term subset from a Phase II trial, a four-patient first-in-human study, news analysis of serious adverse events, and a competition milestone are different evidence classes. This site keeps them labeled so a headline never collapses into “AI cured aging.”