In June 2025, an article published in Nature Medicine — one of the most prestigious scientific journals in the world, with an impact factor of 58.7 — presented the results of the first Phase IIa clinical trial of a drug discovered and designed end to end by artificial intelligence. The drug is called rentosertib (ISM001-055). It was developed by Insilico Medicine to treat idiopathic pulmonary fibrosis (IPF). And it worked.
The conventional drug discovery process — from the identification of a molecular target to entry into clinical studies — typically takes 10 to 15 years and costs hundreds of millions of dollars. Insilico completed this journey for rentosertib in a fraction of that time using its Pharma.AI platform.
The process began with AI identifying the kinase TNIK (Traf2- and NCK-interacting kinase) as a promising target for IPF — a disease with few effective treatments and high mortality. TNIK had not previously been explored as a target for IPF. Then the AI generated and optimized candidate molecules that specifically inhibit this protein. The selected molecule was rentosertib.
The Phase IIa trial involved 71 patients with IPF in a 12-week, double-blind, placebo-controlled study. The group that received 60mg daily of rentosertib showed a change of +98.4 mL in forced vital capacity (FVC) relative to baseline — versus -20.3 mL in the placebo group. A difference of approximately 120 mL, clinically significant in a disease that typically causes progressive deterioration of lung function.
The results were simultaneously published in Nature Medicine and presented at the American Thoracic Society (ATS) Annual Conference in 2025.
Clinical trials fail all the time. The success rate of compounds that enter Phase I and reach approval has historically been 5 to 10%. What makes rentosertib historically significant is not just that it worked — it is that it is the first clinical proof-of-concept record of a drug discovered and designed exclusively by AI, with no participation of human medicinal chemists in the molecule generation process.
This settles part of the debate over whether AI can really contribute in the discovery phase. The experimental answer, now with peer-reviewed published evidence, is: yes.
What is still unresolved: Phase IIb/III (larger, long-duration studies for regulatory approval) still needs to be completed. The FDA granted orphan drug designation to rentosertib in 2023, which accelerates the process. But final approval is estimated for 2027-2028.
In May 2026, Isomorphic Labs — the AI drug discovery company spun off from Google DeepMind — raised US$ 2.1 billion in a Series B round led by Thrive Capital. The round included Alphabet itself, GV, the United Kingdom's sovereign AI fund and other institutional investors.
Isomorphic uses proprietary and more powerful versions of AlphaFold 3 — the protein structure prediction model that won the 2024 Nobel Prize in Chemistry for Demis Hassabis — combined with generative small-molecule design models. The platform, called IsoDDE (Isomorphic Drug Design Engine), designs small molecules to inhibit target proteins from their predicted 3D structures.
Partnerships with Eli Lilly and Novartis (announced in 2024, with a total potential value of approximately US$ 3 billion) and Johnson & Johnson (2025) indicate that large pharmaceutical companies are already betting on this approach. CEO Demis Hassabis confirmed in 2026 that oncology candidates are close to entering Phase I studies with human patients.
The merger between Recursion and Exscientia, completed in November 2024, created the largest AI drug discovery company by pipeline. Recursion processes 2.2 million biological experiments per week across 50 human cell types. Exscientia contributes precision molecular design and more than 60 petabytes of proprietary data.
The combined pipeline has 11 programs, 4 in Phase II. The main setback of 2025 was the termination of the REC-994 program for cerebral cavernous malformation — long-term data did not confirm the efficacy signaled in the early studies. It is a reminder that AI accelerates discovery but does not eliminate the risk of failure in clinical translation.
The field's cautious optimism: with AI, more candidates reach the clinic, faster and with better target selection. The probability of success in each individual trial is still uncertain — but the overall rate of scientific progress is accelerating. The milestone of 2025 is rentosertib. The milestone of 2027 may be the first regulatory approval.
