A drug invented from scratch by software has entered its final round of human trials. Insilico Medicine announced today that rentosertib is now in Phase III testing for idiopathic pulmonary fibrosis, a progressive lung disease that kills most patients within two to four years of diagnosis. This is the last testing stage before a company can apply for regulatory approval.
The reason this matters beyond one company and one disease is what it proves about the process. Rentosertib was not a drug that scientists found and then had software check. The AI identified which biological mechanism to target inside the diseased lung, then built the molecule designed to hit that target. No human came up with either step. The full discovery-to-preclinical-candidate timeline was 18 months. The traditional version of that same work takes three to four years on average.
The Phase II results, published in Nature Medicine in June 2025, gave the field its first real evidence. Patients taking the higher dose saw their lung capacity increase by an average of 98.4 mL over 12 weeks. Patients on placebo lost 20.3 mL over the same period. In a disease where any decline is typically permanent, that is a meaningful gap. The FDA granted rentosertib an Orphan Drug Designation back in 2023, a status that speeds up the review process for diseases affecting small patient populations.
Insilico also ran a parallel Phase IIa trial in the United States, enrolling patients separately from the Chinese study, to build a broader evidence base heading into Phase III. The company already has an inhalable version of the same drug in early safety testing, aiming for a formulation that delivers the drug directly to the lungs at lower doses.
The IPF treatment market is currently worth around $3.5 billion globally and is expected to reach roughly $7 to $8 billion by 2035. The competitive environment just got more active: Boehringer Ingelheim received FDA approval for a new IPF drug in October 2025, the first new approval in over a decade. Rentosertib, if approved, would enter a market with growing physician familiarity and expanding patient demand from aging populations worldwide.
Insilico listed on the Hong Kong Stock Exchange in December 2025, raising approximately $293 million. The company now carries a portfolio of more than 40 programs built using the same AI platform. Between 2021 and 2024, 22 of those programs moved from project start to a preclinical candidate in 12 to 18 months each, compared to the industry average of 2.5 to 4 years. The company has struck licensing deals with Exelixis, Menarini, and others, with a combined deal value potential of $2.1 billion, and counts Sanofi, Lilly, and Takeda among its collaboration partners.
For business operators outside pharma, the thing to track is not this one drug but what a Phase III success or failure would confirm. AI-discovered drugs have shown an 80 to 90 percent success rate in Phase I safety trials, well above the 40 to 65 percent for traditionally discovered compounds. Phase II efficacy is where the numbers align closer to the historical average. Phase III is where the real test happens, with larger patient groups across multiple countries. A clear win here would accelerate AI drug discovery investment across the entire industry. A failure would not kill the field but would put a spotlight on how much of the early promise was speed rather than accuracy.
Insilico reports that none of its AI-designed preclinical candidates have been dropped before reaching human trials, a 100 percent conversion rate. The sample size is still small enough that this could be coincidence. Phase III for rentosertib will go a long way toward answering whether the AI is genuinely better at picking winners, or just faster at producing candidates that still face the same unpredictable odds in late-stage human testing.