AI-Discovered Drugs Are Clearing Phase I Trials at Nearly Double the Usual Rate

--- headline: "AI-Discovered Drugs Are Clearing Phase I Trials at Nearly Double the Usual Rate" slug: "ai-drug-discovery-phase1-success" category: "research" story_number: 13 date: "2026-07-24" ---

# AI-Discovered Drugs Are Clearing Phase I Trials at Nearly Double the Usual Rate

Molecules designed with artificial intelligence are sailing through the first stage of human testing at a rate that would have looked impossible a decade ago. Across the small but growing set of AI-discovered drugs that have reached the clinic, roughly 80 to 90 percent are passing Phase I safety trials — nearly double the 40-to-65 percent that the pharmaceutical industry has come to expect from conventionally discovered compounds. The figure has become a rallying cry for a field that has spent years promising to bend the economics of drug development, and in 2026 it is finally producing clinical data to argue about.

The headline number traces back to a 2024 analysis in Drug Discovery Today by Madura Jayatunga and colleagues at Boston Consulting Group, the first systematic attempt to score how AI-originated molecules actually perform once they reach patients. "In Phase I trials, AI-derived molecules can have a success rate of 80–90%, which is substantially higher success rates than historic averages," the authors wrote. Two years on, later tallies of molecules that had completed Phase I — on the order of two dozen candidates, of which roughly 21 succeeded — have kept the range intact.

The clinical progress is riding a wave of money and speed. AI drug-discovery firms have pulled in more than $2 billion in fresh investment in the most recent cycle, and developers now routinely claim they can move from a validated target to a clinical candidate in 12 to 18 months, rather than the four to five years that traditional programs consume — a compression of roughly 70 percent. Industry counts put more than 3,000 AI-assisted drug programs somewhere in the development pipeline, with some forecasters projecting more than 200 AI-enabled approvals by 2030. The clinically mature slice remains far smaller: the Drug Discovery Today analysis identified about 300 AI-native biotechs worldwide but only 67 AI-discovered molecules actually in clinical trials, roughly 1 percent of the industry's total clinical pipeline.

Named companies anchor the story. China- and U.S.-based Insilico Medicine advanced rentosertib, an anti-fibrotic for idiopathic pulmonary fibrosis widely described as the first drug with both an AI-discovered target and AI-designed structure to reach Phase II, and has reported data from dozens of patients. Recursion, now merged with Exscientia, is carrying multiple clinical-stage programs toward near-term readouts. Isomorphic Labs, the Alphabet-owned spinout built on DeepMind's AlphaFold, has said it intends to put its first AI-designed candidate into human trials — a milestone the company has publicly targeted but, as of mid-2026, not yet crossed.

Why it matters

The temptation is to read an 80-to-90 percent Phase I rate as proof that AI has cracked drug discovery. The evidence does not support that leap, and the researchers who produced the number are the first to say so. Phase I largely measures safety and tolerability in healthy volunteers or small patient groups; it is the stage where good chemistry — precisely what AI is best at optimizing — pays off. The far harder question is whether a drug actually works, and that is answered in Phase II. There, the same BCG team found the advantage evaporates. "In Phase II trials, our data indicate a success rate of AI-discovered molecules of 40%, which is in line with historic industry averages," they wrote — no better than compounds discovered the old way.

That drop-off has a clear mechanistic explanation. AI has dramatically accelerated the design and optimization of molecules, but Phase II failures are overwhelmingly failures of biology: the drug is safe and behaves as designed, yet the target turns out to be the wrong one for the disease. Picking the right biological target remains the field's genuinely unsolved problem, and AI has so far barely touched it. The sample sizes involved are also small enough that a handful of outcomes can swing the percentages, and survivorship bias lurks in any pipeline still too young to have accumulated many late-stage failures.

Even a partial edge, though, compounds. The BCG authors estimated that stacking the observed Phase I and II rates onto historical Phase III performance lifts a molecule's overall odds of clearing all clinical phases from the traditional 5-to-10 percent to roughly 9-to-18 percent — what they called "a near doubling of pharmaceutical R&D productivity." In an industry where the average approved drug costs well over a billion dollars, doubling the hit rate is transformative even if it never touches efficacy.

What to watch

The next 24 months are the real test. With 15 to 20 AI-originated programs expected to enter pivotal or later-stage trials in 2026 and a wave of Phase II readouts due, the field is about to learn whether its Phase I success story survives contact with efficacy data. No AI-designed drug has yet won FDA approval, and that remains the milestone that matters.

Watch three things: whether Insilico's rentosertib and Recursion's lead programs post convincing Phase II efficacy, not just safety; whether Isomorphic Labs actually enters the clinic on schedule; and how regulators frame AI-derived evidence as the FDA builds out its review approach. Analysts at the 2026 J.P. Morgan Healthcare Conference framed clinical trials themselves as the "last untouched bottleneck" — the part of the pipeline AI has yet to meaningfully accelerate. The Phase I numbers are real and encouraging. They are also, for now, a measurement of how well AI designs molecules, not of whether those molecules cure anyone.