A 65-year-old with idiopathic pulmonary fibrosis (IPF) — a lung disease that scars tissue until breathing fails — now has a new kind of hope: a medicine whose disease target was picked by an algorithm and whose chemical structure was drawn by a generative model. In July 2026, that drug, Rentosertib, entered Phase III clinical trials. If it succeeds, it would be one of the clearest proofs yet that AI can do more than write text — it can design medicine.
The developer is Insilico Medicine (英矽智能), a clinical-stage biotech founded in 2014 and headquartered in Hong Kong, also listed on the Hong Kong exchange (HKEX: 3696). Rentosertib, also known as INS018_055 / ISM001-055, is an oral inhibitor of a protein called TNIK, pursued as a treatment for IPF.
Key takeaways
- Rentosertib's target (TNIK) was prioritized by Insilico's PandaOmics AI engine, and its molecule was generated by the company's Chemistry42 generative chemistry platform — both steps AI-driven.
- A Phase IIa trial in 71 IPF patients across 22 sites in China showed the 60 mg-once-daily arm improved lung capacity (FVC) by +98.4 mL over 12 weeks, versus a −20.3 mL decline on placebo; results were published in Nature Medicine (June 2025).
- Insilico initiated a Phase III trial on July 7, 2026, planning to enroll 320 patients over 52 weeks, led by Peking Union Medical College Hospital (北京协和医院).
- Per a Reuters interview (July 2026), Insilico's CEO says its typical path to a developmental candidate is ~13 months versus ~4.5 years by traditional methods — a company-reported efficiency claim, not an audited industry average.
From a keyword to a clinic
The conventional drug pipeline is slow and wasteful: finding a target, then a molecule, then proving it safe and useful can take the better part of a decade and cost billions. Insilico's bet is that AI compresses the front end. Its PandaOmics scans omics data, patents, literature, and aging-related signals to nominate targets; Chemistry42 then designs and refines candidate molecules. Rentosertib is the headline output of that stack.
The Phase IIa readout, published in Nature Medicine on June 3, 2025, was the field's first clinical proof-of-concept for an AI-discovered drug. In the GENESIS-IPF study (registry NCT05938920), 71 patients in China were randomized to placebo or one of three Rentosertib doses for 12 weeks. The highest dose produced a dose-dependent improvement in forced vital capacity (FVC), the standard lung-function measure, while placebo declined. Biomarker analysis showed reductions in profibrotic proteins, consistent with the intended mechanism. The trial was small, and the authors were careful: larger, longer studies were needed.
The Phase III leap
On July 7, 2026, Insilico announced the Phase III trial — a randomized, double-blind, placebo-controlled study of 320 patients over 52 weeks. The lead principal investigator is Professor Zuojun Xu (徐作军) of Peking Union Medical College Hospital, with Academician Nanshan Zhong (钟南山), a towering figure in Chinese respiratory medicine, and President Chang Chen (陈昶) of Shanghai Pulmonary Hospital as co-leads. The drug had already received U.S. FDA orphan-drug designation in February 2023, reflecting IPF's unmet need.
Why China matters here: the proof-of-concept and the ongoing late-stage testing are anchored in Chinese clinical sites, and Insilico's experimental validation and scaling are concentrated in the country. The company argues that combining Chinese wet-lab scale with generative design is a dual accelerator — a claim about operational speed, not about any single molecule being approved faster.
A credible, bounded claim
Insilico's founder and CEO, Alex Zhavoronkov, told Reuters in July 2026 that the company typically reaches a developmental candidate in about 13 months, with a record of 9 months, against a traditional benchmark of roughly 4.5 years, and that AI has helped generate 31 developmental candidates over six years. Those are the company's own averages, reported via Reuters; they are not regulator-audited metrics and exclude Phase III success rates or total capital to approval. They are best read as evidence of a faster early pipeline, not a guaranteed faster drug.
Honest limitations
This is one program, not a verdict on AI drug discovery as a whole. The Phase IIa signal was encouraging but from a small, exploratory trial; the −20.3 mL placebo decline and +98.4 mL treated change are point estimates with confidence intervals that overlap at the edges, and the authors themselves called for validation in larger cohorts. Phase III has not reported, so safety and efficacy at scale remain unknown. The efficiency figures are company-reported and self-selected. IPF affects roughly 5 million people globally, but a single asset does not represent the breadth of AI-discovered pipelines. I have not assessed Insilico's financials or other pipeline assets beyond what is needed to frame this milestone, and "first AI-designed drug" claims depend on definitions — I describe Rentosertib as a leading, fully AI-designed candidate rather than asserting an absolute first.
What readers can do now
- If you follow biotech, treat Rentosertib's Phase III as a real but early test of whether AI-designed drugs can clear the hardest clinical bar — watch the 52-week readout, not the press release.
- For patients and families dealing with IPF, this is research-stage; discuss registered therapies and clinical-trial options with a specialist rather than treating any report as a treatment available today.
- For builders in AI, the takeaway is methodological: AI's value here is generating and ranking hypotheses (targets, molecules) that still require rigorous, regulated experimentation — the model proposes, the lab disposes.