AI-Discovered Drug Rentosertib Begins 320-Patient Phase III Trial

A milestone that comes with an asterisk

AI drug discovery now has its first real late-stage test. Insilico Medicine dosed the first patient in GENESIS-IPF-3 on September 9, 2026. The Phase III trial is testing rentosertib, an oral pill for idiopathic pulmonary fibrosis (IPF), and the first doses were given at Peking Union Medical College Hospital, with Shanghai Pulmonary Hospital enrolling its first patient the same day.4 The company calls rentosertib the first drug in which both the disease target and the molecule were produced by AI to reach Phase III.4 Coverage has mostly accepted that label without much pushback.

The trial is randomized, double-blind and placebo-controlled. It aims to enroll 320 patients at 47 centers in China, and patients take the drug once a day for 52 weeks.7 The main measure is how quickly forced vital capacity (FVC), the amount of air a patient can forcefully exhale, declines over a year. The key secondary measure is time until the disease first gets worse.23 Nobody has efficacy data yet. The drug is still investigational and no regulator has approved it.23

The timeline in the reporting is a little confusing. Insilico announced on July 7 that it had "initiated" the Phase III, and trade outlets wrote about it the next day.91 The first patient was not dosed until two months later.6 Some articles treat July as the start and others use September. September is the date that counts, because that is when patients actually began receiving the drug.

Why the target matters more than the molecule

Rentosertib blocks TNIK, a kinase that Insilico's PandaOmics software ranked as a top fibrosis target by combining multi-omics data from fibrotic tissue with scores tied to aging biology.25 The company's generative chemistry platform, Chemistry42, then designed the molecule.1 Zuojun Xu, the trial's lead investigator, said TNIK had never been linked to fibrosis before. He described that as a sign that AI may be finding targets in ways traditional research does not.27

This is the important point, and some of the coverage misses it. Plenty of computational tools already speed up chemistry. The bigger claim is that an algorithm found a biological target humans had overlooked, and that the target actually works in patients. One analysis framed the trial as the field's first real exam: a positive result would show AI can pick targets that beat placebo, while a failure would suggest AI mainly makes discovery faster without producing better drugs.21 That seems right. Rentosertib works by a mechanism that neither pirfenidone nor nintedanib uses. Those are the two antifibrotics the FDA approved in October 2014, and both slow the disease without stopping it.24 A new mechanism raises the potential payoff. It also leaves less earlier evidence to fall back on.

What the earlier data does and doesn't show

The move to Phase III rests on a small, short study. In the Phase IIa trial, published in Nature Medicine in June 2025, patients taking 60 mg once a day gained an average of 98.4 mL in FVC over 12 weeks, while the placebo group lost 20.3 mL.25 That study met its main safety goal, and side-effect rates were similar across all arms.3 It also enrolled only 71 patients at 22 sites and lasted 12 weeks, which is far less evidence than the Phase III is designed to produce.25

That difference is the main risk. A lung-function improvement over 12 weeks in a few dozen patients is promising. A slower rate of decline over a full year in 320 patients is the kind of evidence regulators need. One review noted that AI-designed drugs still fail at Phase II about as often as conventionally discovered ones.22 Another writer, putting odds on the outcome, estimated a 40% chance that a drug with an AI-chosen target and AI-designed molecule wins U.S. approval by 2030. That writer pointed to the China-only design, the missing readout date, and earlier AI-drug results that did not hold up.21

The outlets also differ on how much to say about aging. One consumer-tech site said rentosertib "might reverse your biological age," citing preclinical work.26 Insilico's own July release was more careful. It said its senescence findings strengthen the rationale for studying TNIK but do not make rentosertib an anti-aging therapy.3 In this case the company's caution is the better guide.

The timeline nobody can pin down

According to coverage of Xu's comments and the company's own guidance, approval could come three to four years after Phase III began, if things go well.2310 Because each patient is treated for 52 weeks, one trial-industry editor concluded that the main results are unlikely before late 2027 at the earliest.24 Other observers simply list the readout as undated or put it in "2027 and beyond."2122

Geography is another open question. All 47 sites are in China.7 The FDA granted rentosertib Orphan Drug Designation in February 2023, and China's Center for Drug Evaluation granted Breakthrough Therapy Designation in May 2025.25 It is not yet clear whether a trial run only in China will be enough for a U.S. filing, and that is the main reason U.S. approval estimates remain cautious. Insilico is also testing an inhaled version in a Phase I study cleared in April 2026, though it has not released any data comparing the two forms.25

"First" depends on your definition

The "first AI-discovered drug" headline is narrower than it sounds. The claim holds only for drugs where AI produced both the target and the molecule. By broader definitions, other AI-linked drugs are as far along or further. Generate:Biomedicines announced two Phase III trials of its AI-engineered antibody GB-0895 for severe asthma in December 2025, covering about 1,600 patients in more than 40 countries.22 Relay Therapeutics has its PI3Kα inhibitor zovegalisib in Phase III for breast cancer.28 Zasocitinib, from Nimbus and Schrödinger's physics-plus-machine-learning approach and now owned by Takeda, is under FDA review. One overview suggests it could become the first AI-designed drug approved, possibly in early 2027.22

So rentosertib is probably not the first AI-linked drug to reach patients after approval. If it succeeds, it would be the first proof that AI can find a disease target on its own, which is a different and more ambitious claim.

Capital has run ahead of evidence

Investors are not waiting for the Phase III results. Isomorphic Labs, the DeepMind spinout built around AlphaFold-style protein structure prediction, raised a $2.1 billion Series B in May 2026. Thrive Capital led the round, with Alphabet, GV, MGX, Temasek, CapitalG and the UK Sovereign AI Fund also taking part.11 That brings its disclosed outside funding to about $2.7 billion.13 Its first clinical trials were expected by the end of 2026, but it has not published human data that compares with Insilico's.2213 One market review put it plainly: the best-funded platform is not necessarily the one closest to an approved drug.13

The totals vary by source. One tally says AI drug-discovery startups raised more than $5.2 billion in 2025–2026, and that 173 AI-originated programs are now in clinical trials.12 The same source credits Xaira with $1.3 billion,12 while others describe its 2024 launch funding as about $1 billion.14 Different methods produce different numbers, but they agree on the overall picture: very large sums are going to companies with little or no clinical data.

Insilico pays for its work differently. The company has listed out-licensing and co-development deals with Eli Lilly, Servier, Takeda and others as the basis of its progress.27 One editor pointed out that funding a Phase III with licensing income rather than new shares could change how long Insilico can keep going, and suggested watching that alongside the FVC results.24 The company went public on the Hong Kong Stock Exchange on December 30, 2025.27

Protein design is the next test

Rentosertib is a small molecule. The next wave of the field is designing proteins from scratch, and none of those programs has reached human testing. Xaira was co-founded by Nobel laureate David Baker and is led by former Genentech research chief Marc Tessier-Lavigne. In October it described its X-Design antibody platform and gave the first details of its pipeline. XA-1, a cancer candidate, went from start to lead molecule in seven weeks. XA-4 targets a G-protein coupled receptor that the company described as "seemingly intractable."31 In its technical write-up, Xaira said conventional library screening and llama immunization both failed against XA-4, while its AI-designed antibodies produced a working, cross-reactive antagonist.32 The company also says that having good leads does not guarantee clinical success, and Tessier-Lavigne would not give a timeline for human trials.3231

That leaves the field at an uneven point. Structure prediction and generative protein design are improving fast in the lab and drawing large amounts of funding. The question that matters most to patients, whether an AI-found target can slow lung scarring better than a placebo, will be answered by a 52-week trial in 320 people. Until those results arrive, the claims about AI drug discovery have not been proven in late-stage trials.