ena pragma
·6 min read·ai-verification · ai-market · ai-drug-discovery

The AI drug 'reversing aging' was never tested for that. A different AI drug already has conditional approval.

An AI summary collapses three separate papers about one trial into one, and misses the AI-assisted drug that already holds a conditional market approval.

Contents

A Google AI Mode summary on AI, longevity medicine and human tissue tells a simple story: AI designs a drug, and the drug reverses aging. The real record is three separate papers about one trial, plus a second drug the summary never mentions at all, an AI-assisted one that already holds a conditional market approval. Neither correction is a technicality. Both change what a reader should actually take from this.

#What the trial actually tested, and what it found

Rentosertib (formerly ISM001-055) is a first-in-class, AI-generated small-molecule inhibitor of TNIK, a target for idiopathic pulmonary fibrosis (IPF) discovered by Insilico Medicine's generative-chemistry platform. Its Phase 2a trial (NCT05938920, 71 patients across three dose arms and placebo) had one primary endpoint: the percentage of patients with at least one treatment-emergent adverse event, which came back similar across every arm, 70.6 percent on placebo against 72.2 to 83.3 percent on the three dose arms. That is a safety readout, not an efficacy claim, and it is the number the trial was actually designed to answer.

Lung function, measured as forced vital capacity (FVC), was a secondary endpoint. At 60 mg once daily the trial reported a mean FVC change of +98.4 mL (95% CI 10.9 to 185.9) over 12 weeks, against -20.3 mL (95% CI -116.1 to 75.6) for placebo. That is a real, reported number from a real trial, in a small, early-phase study whose own authors describe the result as warranting "further investigation in larger-scale clinical trials of longer duration," not as an established treatment effect. A Phase 3 trial is now underway; its result is not expected before 2027.

#The three papers the summary collapsed into one

The AI Mode page attributes the Phase 2a trial to Nature Biotechnology. It was published in Nature Medicine. That single misattribution is a symptom of a bigger compression: Insilico's own account of this work spans three distinct papers, published in two different journals across three years, and the summary treats them as one.

The discovery-to-clinic story, how the target and molecule were found, ran in Nature Biotechnology in 2024. The Phase 2a trial results above ran in Nature Medicine in 2025. A third, separate paper, published in Nature Biotechnology in 2026, took stored blood serum from that same 2025 trial and reanalyzed it with six independent proteomic aging clocks, tools that estimate a person's biological age from protein markers rather than their calendar age. This third paper is where the "reversed biological age" language comes from, and it was neither a pre-specified endpoint of the original trial nor a claim about the drug treating aging itself. Insilico's own press release on this reanalysis states the finding plainly: researchers reported the largest effect at week 4 in the group receiving 30 mg twice daily, with the clocks estimating roughly 3 to 4 years of difference from placebo, and up to 6 years on one individual clock. That is what six aging clocks measured in a secondary, exploratory reanalysis of existing samples, reported by the company that ran the trial. It is not a clinical claim that the drug reverses aging, and no regulator has evaluated it as one.

#The AI-assisted drug that already has conditional approval, and the summary never mentions

While one AI-discovered molecule moves through Phase 3 trials for a chronic disease, a different AI-assisted drug already holds a conditional market approval, and the AI Mode summary does not mention it once. China's National Medical Products Administration granted conditional market approval in late July 2026, roughly five weeks before the coverage that reported it, to Mprosevir, developed by Westlake University and Westlake Pharmaceuticals for mild-to-moderate COVID-19. (Drug-intelligence records put the approval at July 28; the manufacturer's supply-chain partner says July 29. The news coverage carries a September dateline and no event date at all, which is its own small version of the problem this piece is about.) It is reported as China's first AI-assisted Class 1 innovative drug and the first small-molecule drug approved anywhere based on DNA-encoded library (DEL) screening technology.

The AI's role here was different in kind from rentosertib's. DEL screening tested a library of 49 billion compounds against a known target class, an AI model narrowed the initial hit list of more than 100 candidates down to nine for physical testing, and six of those nine showed strong activity. That is AI-accelerated screening of a known approach, not the invention of a new molecule and target from nothing. The often-repeated "three and a half years" figure describes discovery to the completion of clinical trials, not discovery to regulatory approval, which came afterward.

#What actually differs, and what does not

These are two different uses of AI in drug development, at two different regulatory stages, and treating them as one undersells the more interesting comparison. Rentosertib is AI inventing a molecule and a target that did not exist before, still in trials, with the age-related finding sitting outside the trial's own primary purpose. Mprosevir is AI accelerating a known screening method against a known disease target, already cleared for conditional market use. Neither case changes the evidence bar a drug has to clear. A safety endpoint is still a safety endpoint, a secondary finding is still secondary, and a conditional approval is still conditional, no matter which of these two approaches produced the candidate molecule. What is genuinely new sits in the discovery stage: on the Mprosevir program, narrowing more than a hundred candidates to nine took a matter of days. The strength of the evidence once a candidate reaches trials is unchanged. A reader evaluating any "AI just did X" claim in medicine should ask the same question of both stories: what was the endpoint actually designed to measure, and is the number in the headline that endpoint, or something a paper found afterward while looking somewhere else. A correctly quoted number and the interpretation built on top of it are two separate claims, and the only way to catch a gap between them is opening the primary paper rather than trusting an AI-generated summary of it, which is exactly how a real trial about lung safety became a headline about reversing age.

#Sources

Keep reading