AI-designed drugs are closing in on their first regulatory approval
Generative biology has quietly moved from research demo to clinical reality. As of mid-2026, more than 173 drug programs whose molecules were designed by AI are in active clinical trials — up from 67 in 2023 and just 3 in 2016. The field's furthest-along candidate, Insilico Medicine's rentosertib for idiopathic pulmonary fibrosis, is now in Phase III trials, and analysts put roughly 60% odds on the first fully AI-designed drug winning regulatory approval sometime in 2026 or 2027.
This matters because drug discovery has historically been the slowest, most expensive part of bringing new medicine to patients — a process that traditionally takes 10-15 years and over a billion dollars per approved drug, with most candidates failing along the way. AI-designed drugs aren't skipping clinical trials or regulatory scrutiny, but they are compressing the front end of that pipeline dramatically, and the trial data now emerging is the first real test of whether AI-generated molecules perform as well in humans as traditionally discovered ones.
How Generative Biology Actually Works
Unlike traditional drug discovery, which screens existing molecule libraries or makes incremental modifications to known compounds, generative biology models design entirely novel protein structures and small molecules from scratch, optimized computationally for a specific biological target before any wet-lab synthesis happens. Insilico Medicine's platform, for example, uses generative AI to both identify a disease target and design a novel molecule against it — a process the company says can compress the traditional multi-year discovery phase into months.
Rentosertib, Insilico's lead candidate, targets fibrosis — a disease process where AI has particular strengths, since fibrotic tissue remodeling involves complex multi-protein interactions that are difficult to model with traditional structure-activity approaches. In Phase IIa results reported in mid-2025, the drug showed measurable improvement in lung function among idiopathic pulmonary fibrosis patients, a disease with few effective treatment options and a five-year survival rate worse than many cancers.
Who's Actually in the Clinic
Insilico's pipeline is the deepest in the field: rentosertib in Phase III, three additional programs in Phase II, and eight in Phase I. But the company isn't alone. Generate Biomedicines is designing novel protein therapeutics computationally from the ground up rather than modifying existing ones. Recursion Pharmaceuticals, following its merger with Exscientia, runs a vertically integrated AI discovery platform with several active clinical candidates — Exscientia previously took an AI-designed small molecule for obsessive-compulsive disorder from concept to human trials in just 12 months, a fraction of the traditional timeline. BenevolentAI has BEN-8744, a PDE10 inhibitor for ulcerative colitis, in Phase I. Relay Therapeutics is running Phase I trials on RLY-1971, an SHP2 inhibitor for solid tumors.
Industry-wide, 15 to 20 AI-originated drug programs are expected to enter pivotal Phase III trials during 2026 alone — the stage immediately preceding an approval filing. That volume, arriving simultaneously across multiple companies and disease areas, is the strongest signal yet that generative biology has moved past proof-of-concept into a repeatable drug discovery methodology.
Why Regulators Are Racing to Keep Up
The FDA has been building regulatory infrastructure specifically for AI-originated drugs rather than treating them as a special case within existing pathways. The agency launched its CDER AI Pilot Program and, in December 2025, qualified its first AI tool for use directly in clinical trial design and analysis. The agency's public position emphasizes that AI-designed drugs will be held to the same evidentiary standard as any other candidate — robust trial data and human oversight of the AI-driven claims — rather than getting an accelerated approval pathway simply because AI was involved in discovery.
That distinction matters for how the industry should read the "first approval by 2026-2027" forecasts: the AI acceleration happens almost entirely in the discovery and design phase, not in the clinical trial phase. A drug still needs to clear the same Phase I, II, and III safety and efficacy bars as any traditionally discovered compound. What's changing is how fast a promising molecule reaches the starting line of that process, not how fast it clears the finish line.
What Comes Next
In May 2026, the nonprofit Biohub released an open-source AI model for protein design aimed initially at cancer and immune-system targets — a move that could lower the barrier to entry for smaller biotech startups that can't build proprietary generative biology platforms from scratch. If open models produce clinical-grade candidates, expect the current field of five or six well-funded leaders to expand rapidly, similar to how open-source large language models multiplied the number of companies building AI products once the underlying model layer became accessible.
For biotech investors and pharma partnerships, the practical signal to watch is Insilico's Phase III readout for rentosertib — a positive result would be the strongest evidence yet that AI-designed molecules can match traditional discovery's clinical success rates, and would likely trigger a wave of licensing deals and partnership announcements from larger pharma companies that have so far mostly watched generative biology from the sidelines rather than committing capital to it directly.