An AI Built This Drug From Scratch. Now It Faces Its Biggest Test.
Insilico Medicine's rentosertib just became the first drug fully discovered by AI to enter a pivotal Phase 3 trial. The lung disease study could validate an entire technology, or remind everyone that biology doesn't care who designed the molecule.
The Robot Scientist Goes to Phase 3
Back in 2020, the very first molecule designed by artificial intelligence entered a human clinical trial. It was a curiosity, a science fair project with a pulse. Six years later, the experiment has grown up.
Insilico Medicine just announced that rentosertib, a drug discovered entirely by generative AI, has entered a Phase 3 clinical trial for idiopathic pulmonary fibrosis (IPF). That's the scarring lung disease that slowly suffocates patients over years, with no cure and limited options. If this trial works, it won't just be a win for lung patients. It'll be the strongest evidence yet that AI can do what pharma has spent decades and billions of dollars trying to do: find good drugs, faster.
But "if" is doing a lot of heavy lifting in that sentence.
What Makes This Drug Different
Plenty of companies slap the "AI-powered" label on their work. Sometimes it means a computer helped screen a library of molecules. Sometimes it means an algorithm picked a target. Rentosertib is different because AI did both. Insilico's platform, called Pharma.AI, identified the biological target and then designed the molecule from scratch. No human chemist sketched the first version on a whiteboard.
Think of it like the difference between using GPS to navigate a road trip versus having the GPS plan the route, build the car, and pick the destination. Rentosertib is the second version.
The company says it went from target discovery to clinical candidate in under 30 months. For context, that process typically takes four to six years in traditional pharma. That speed alone has made Insilico a poster child for the AI drug discovery movement.
The Trial That Will Settle the Debate
The Phase 3 study is a textbook design: 320 patients across 47 centers in China, randomized to either rentosertib or placebo, taken once daily for 52 weeks. The primary endpoint is the annual rate of decline in FVC, which is a measure of how much air your lungs can push out in a single breath. In IPF, that number drops steadily. A good drug slows the drop.
Get tomorrow's biotech intelligence before your competitors.
Join thousands of biotech professionals who start their day with our free, daily briefing.
The approved IPF treatments, including nintedanib (Ofev), pirfenidone (Esbriet), and nerandomilast (Jascayd), do exactly that: they slow the decline. They don't reverse it. They don't cure it. And they come with significant gastrointestinal side effects that make some patients stop taking them altogether.
That's where the unmet need sits. The global IPF treatment market is roughly $4.1 to $5.4 billion in 2026, and it's growing because patients still don't have anything close to a satisfying answer. If rentosertib can match or beat the existing drugs on lung function while being easier to tolerate, it has a real commercial shot.
Early Data: Promising but Small
Before you get too excited, let's look at what we actually know. The Phase 2a trial, called GENESIS-IPF, enrolled 71 patients at 22 sites in China. It was a 12-week study designed primarily to test safety.
On that front, the drug passed. Side effects were similar across all groups, and most were mild to moderate. The main reasons patients dropped out were liver toxicity and diarrhea, which is worth watching but not unusual in this space.
The efficacy signal was the headline grabber. Patients on the 60 mg once-daily dose saw their lung function improve by an average of +98.4 mL, while placebo patients declined by -20.3 mL. That's a meaningful gap in a disease where lungs are supposed to only get worse.
But caveats matter here. The study was small, short, and conducted only in China. Twelve weeks is not fifty-two weeks. Seventy-one patients is not three hundred and twenty. The Phase 3 trial exists precisely because the Phase 2a results, while encouraging, need to be confirmed in a bigger, longer, more rigorous setting.
The Bigger Picture: AI's Coming-of-Age Story
Rentosertib isn't just a drug; it's a referendum on an entire technology. By 2023, there were 67 AI-derived drugs in clinical trials worldwide. Today, depending on how loosely you define "AI-discovered," there are somewhere between 75 and 173 programs in the clinic.
Insilico itself has been on a tear. The company says it nominated nine preclinical candidates in the first nine months of 2026 alone, pushing its total pipeline to 31 candidates across oncology, metabolism, inflammation, pain, and fibrosis. That pace would be impressive for a mid-size pharma company with thousands of chemists. For an AI-first platform, it's a statement.
Other AI-designed molecules have also reached Phase 3, including an antibody called GB-0895. But rentosertib gets the most attention because its AI pedigree is so thoroughly documented, from target to molecule to clinical trial, all in peer-reviewed publications.
The "Yeah, But" Section
Let's be honest about what this milestone does and doesn't prove.
What it validates: AI can identify a novel target, design a molecule against it, and produce a candidate good enough to reach pivotal testing. The discovery timeline was genuinely compressed. That's real.
What it doesn't validate (yet): that AI-discovered drugs are more likely to succeed in Phase 3, or that they work better than conventionally discovered drugs, or that any of this will lead to a faster FDA approval. Phase 3 trials fail roughly 40-50% of the time across the industry. AI doesn't get a special pass.
The clinical question is still wide open. Will rentosertib slow lung function decline over a full year? Will the safety profile hold up in a larger population? Will it work outside of China? None of those answers exist yet.
Why You Should Care Anyway
Even skeptics should pay attention to this trial. IPF patients desperately need better options; the current drugs slow the disease but don't stop it, and the side effects are rough. A new mechanism that shows real benefit would change the standard of care for a disease that affects millions globally.
And for the broader biotech ecosystem, the stakes are even bigger. If rentosertib succeeds, it will unlock a wave of investment and confidence in AI drug discovery. If it fails, it won't kill the field (the pipeline is too deep for that), but it will temper the hype in a way that the industry's boosters aren't prepared for.
For now, rentosertib sits in the most exciting and most uncertain place in medicine: a Phase 3 trial with something to prove. The robot scientist built the drug. Now it has to work on actual humans, for an actual year, against an actual disease. No algorithm can guarantee that part.
Clinical & Regulatory4 min read
Lutathera Just Got Its First Copycat. Radioligand Therapy Will Never Be the Same.
The FDA just approved the first-ever generic radioligand drug, a copycat of Novartis' blockbuster Lutathera. It's a regulatory first that could crack open pricing and competition in one of oncology's most complex (and expensive) drug categories.
