Insilico Medicine, the Hong Kong- and New York-based biotech that has spent nearly a decade building AI systems to invent new drugs from scratch, hit two regulatory and clinical milestones within three weeks of each other this summer. On July 7, the company’s rentosertib — a TNIK-inhibitor designed by its generative AI platform for idiopathic pulmonary fibrosis (IPF), a progressive and ultimately fatal scarring of the lungs — entered a Phase 3 trial, making it, by most industry trackers’ accounting, the furthest-advanced drug ever to have been both discovered and designed by artificial intelligence. Then, on July 29, the FDA granted Fast Track Designation to a second Insilico compound, ISM6331, a pan-TEAD inhibitor being developed for advanced mesothelioma, a rare and aggressive cancer most often linked to asbestos exposure.
A Company Built Around One Bet: Let the Algorithm Pick the Target
Insilico has staked its identity on an end-to-end AI pipeline — tools it brands PandaOmics for identifying disease targets and Chemistry42 for generating novel drug-like molecules — rather than applying AI to just one step of the traditional discovery process. Rentosertib, originally designated ISM001-055, was the company’s proof of concept: both the biological target (TNIK, a kinase implicated in fibrotic signaling) and the specific molecule were nominated by its AI systems, a claim the company has repeated in investor materials and academic presentations for several years as its flagship example of generative drug design actually reaching the clinic rather than staying in a slide deck.
Why IPF Was the Right Proving Ground
Idiopathic pulmonary fibrosis has for years been considered an area of high unmet need: only two antifibrotic drugs, pirfenidone and nintedanib, had been FDA-approved for the disease prior to rentosertib’s trial, and both slow progression without reversing damage or dramatically extending survival for most patients. That gap — a well-defined disease, a clear biological rationale around fibrotic signaling, and a thin competitive field — made IPF an attractive first target for Insilico to demonstrate that an AI-originated molecule could survive Phase 1 and Phase 2 scrutiny before reaching the much larger, more expensive Phase 3 stage now underway.
The Optimists’ View: Proof the Model Works
Insilico and its backers argue that rentosertib’s advance to Phase 3, paired with ISM6331’s Fast Track designation for mesothelioma, demonstrates that AI-driven discovery is no longer theoretical — it is producing molecules that regulators and investigators treat the same as any other promising candidate. Fast Track status, which the FDA grants to therapies addressing serious conditions with unmet medical need, also means ISM6331 is now eligible for more frequent FDA interactions and potential rolling review, which can meaningfully shorten the path to approval if trial data holds up.
The Caution: Phase 2 Failure Rates Haven’t Budged
Despite milestones like these, a broader look at the field tempers the enthusiasm. An analysis presented at the American Society of Clinical Oncology’s 2026 meeting counted 117 AI-enabled therapeutic assets in human trials across 63 companies, but found only 8 of them — fewer than 7% — had completed Phase 2 testing, the stage where many drugs of any origin fail. A separate Nature Medicine analysis cited across the industry this year found that AI-discovered drugs have so far failed in Phase 2 at rates statistically similar to conventionally discovered compounds, directly challenging the pitch that AI-originated molecules are inherently more likely to work once tested in people. Rentosertib reaching Phase 3 is a genuine milestone, but it is also, by definition, a single data point in a field still waiting for its first full regulatory approval.
What’s Next
As of this reporting, no AI-discovered or AI-designed drug has received full FDA approval — rentosertib’s Phase 3 trial and ISM6331’s Fast Track pathway are both still mid-journey rather than finished stories. The coming 12 to 24 months of trial readouts will matter far beyond Insilico’s own balance sheet: a Phase 3 success for rentosertib would be the strongest evidence yet that generative AI drug discovery can consistently clear the same bar as traditional pharmaceutical R&D, while a failure would hand ammunition to skeptics who argue the field’s milestones have outpaced its actual clinical track record.
Photo: Kindel Media / PEXELS via Pexels
