The Cocktail Report (sounds really smart around your friends):
Artificial intelligence picked the target and designed the molecule: rentosertib, a first-in-class blocker of TNIK, a signaling enzyme flagged by software as touching six of the recognized hallmarks of aging.
In 42 lung-fibrosis patients given the drug for 12 weeks, all six independent proteomic aging clocks read the treated groups as biologically younger, with the strongest arms landing 2.7 to 3.5 years below baseline at week four.
The obvious objection fails a direct test: the dose that most improved lung function showed the weakest aging signal, and lung function explained only about 6% of the variation in clock scores.
The drug is senomorphic rather than senolytic, meaning it quiets what worn-out cells secrete instead of killing them, a distinction most television coverage got wrong.
The reality check: AI-designed molecules pass Phase 1 safety at 80 to 90% against a historic 40 to 65%, but pass Phase 2 efficacy at roughly 40%, which is merely average.
If you are waiting to learn whether artificial intelligence can produce actual medicine instead of press releases, your answer starts arriving this week. A drug that AI both targeted and designed just moved six separate measures of biological aging in human patients, then entered the first Phase III trial ever run on a machine-discovered compound.
The drug is rentosertib. It blocks TNIK, a signaling enzyme that Insilico Medicine's software flagged as sitting inside six of the recognized hallmarks of aging, and no approved drug works this way.
They tested it in idiopathic pulmonary fibrosis, a fatal scarring of the lungs that typically surfaces around age 65. Forty-two patients in a randomized, placebo-controlled trial gave blood at four points across 12 weeks.
Researchers measured 2,841 proteins in that blood and ran the results through six proteomic aging clocks, which are machine-learning models that estimate biological age from circulating protein levels rather than from a birth date. All six pointed the same way.
Treated patients came out younger. On the 60mg once-daily dose, the four clocks trained on chronological age placed patients 2.7 to 3.5 years below their own starting point by week four, while the placebo group drifted slightly older.
Now the part that separates this from a company announcement. The obvious objection is that repairing a diseased lung would shift blood proteins that aging clocks happen to read, making a sick patient score younger without anything aging-related occurring.
The authors tested that objection and it did not survive intact. The dose producing the best lung-function gains showed the least consistent aging signal, and lung function accounted for only about 6% of the variation in clock scores.
They also checked the drug's protein shifts against normal aging in 55,319 UK Biobank participants. In the twice-daily group, those protein shifts ran against the direction of aging, while the placebo group's ran along with it.
The mechanism deserves precision, because television coverage has already mangled it. Rentosertib does not clear out senescent cells, the worn-out cells that quit dividing but keep leaking inflammatory signals.
It is senomorphic, which means it silences what those cells secrete and leaves them where they are. The placebo group's senescent-cell signal climbed across the 12 weeks, and every treated group's fell.
Here is the part that keeps this from being a victory lap. You have likely been told that AI is on the verge of curing aging, and the real record says something narrower and more useful than that.
AI-designed molecules clear Phase 1 safety trials at 80 to 90%, against a historic industry rate closer to 40 to 65%. Machines are measurably better at designing compounds that do not hurt people.
Those same molecules clear Phase 2, where a drug must prove it actually works, at roughly 40%. That is the ordinary industry average, and the first wave of AI-designed drugs mostly died there.
The limits on the aging result are real. Forty-two patients means nine to eleven people per group at each timepoint, everyone was an older Asian adult with a fatal lung disease, and the statistical threshold was looser than the customary standard.
The effect peaked at week four and then flattened for the remaining eight. The single protein contributing most to the clock scores was LTBP2, a controller of the lung's own scarring process rather than an aging protein.
The authors state plainly that they cannot separate the anti-fibrotic effect from an anti-aging one, and that resolving it requires testing people who do not have pulmonary fibrosis. The first author is the company's chief executive, though the senior author is a Harvard aging biologist.
Why Should You Care?
The 320-patient, 52-week Phase III that began dosing this week is the first real referendum on whether machine-designed drugs work, not merely whether they are safe. Its answer will set the pace for everything else arriving in longevity medicine.
The quieter finding may matter more. The same total daily dose, split into two smaller ones, produced a different aging signature than a single large dose, which suggests timing is part of the drug and not just chemistry.
1. Zhavoronkov A, Galkin F, Chen S, Ren F, et al. "Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment." Nature Biotechnology, September 7, 2026. https://doi.org/10.1038/s41587-026-03286-y
2. Xu Z, Ren F, Wang P, Cao J, et al. "A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial." Nature Medicine 31(8):2602-2610, June 3, 2025. https://www.nature.com/articles/s41591-025-03743-2
3. Insilico Medicine. "Insilico Medicine Doses First Patient in GENESIS-IPF-3, the World's First Phase III Trial of a Generative AI-Driven Innovative Drug." September 10, 2026. https://insilico.com/news/isn1009261-insilico-medicine-doses-first-patient-genesis-ipf-3
4. Jayatunga MKP, Ayers M, Bruens L, Jayanth D, Meier C. "How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons." Drug Discovery Today, June 2024. https://pubmed.ncbi.nlm.nih.gov/38692505/
5. Endpoints News. "After years of hype, the first AI-designed drugs fall short in the clinic." https://endpoints.news/first-ai-designed-drugs-fall-short-in-the-clinic-following-years-of-hype/
