MIRA diagnosed real emergency room cases more accurately than four board-certified physicians, 87.8% to 78.1%, in a study published in Nature. Days later, a drug an AI designed from scratch entered the final stage of human trials.
Jakob Kather's team at TU Dresden built an AI agent, ran it against real hospital records, and it beat the doctors. Not narrowly, and not on a simplified quiz. MIRA, short for Medical Intelligence for Reasoning and Action, worked through 574 real emergency department cases inside a sandboxed electronic health record, taking patient histories, ordering labs and imaging, interpreting the results, and writing up treatment plans the way a resident would. In a matched comparison against four board-certified physicians on 311 of those cases, MIRA hit 87.8% diagnostic accuracy. The doctors managed 78.1%. The results, from Kather's group at the Else Kröner Fresenius Center for Digital Health alongside Heidelberg University Hospital, ran in Nature on June 17.
You don't get numbers like that from a chatbot answering trivia. MIRA had to navigate what the researchers call a clinical action space: request the right test, read it correctly, decide what it means, and act on it, the same sequence a doctor runs through under time pressure. It wasn't uniformly dominant. On appendicitis, MIRA nailed 98.6% of cases. On pneumonia, it dropped to 72.4%, worse than its own average - the agent still has real blind spots. The system also stayed text-only. No physical exam, no tone of voice, no images read by eye. The cases came from clean, complete datasets, not the chaos of a real overnight shift with an incomplete chart and a patient who can't quite explain what's wrong.
That caveat matters more than it might sound. A lot of the loudest AI health headlines this year have come from benchmarks, tidy multiple-choice tests where a model can look brilliant without ever touching a real workflow. MIRA is a step past that, sitting inside an actual EHR structure and making the kind of sequential calls a hospital shift is built from. It's still a simulation. Nobody let MIRA touch a live patient. But simulation built from real ED cases is a meaningfully different claim than a benchmark built from exam questions, and it's why this result traveled further than most.
Three weeks after that paper landed, Insilico Medicine moved its own AI project into a different kind of proof. On July 7, the Hong Kong-listed biotech announced it had begun a Phase III trial for rentosertib, an oral TNIK inhibitor for idiopathic pulmonary fibrosis, a lung disease that scars tissue until breathing becomes impossible. What makes rentosertib different from other drugs is where it came from: both the TNIK target and the molecule itself were generated by Insilico's Pharma.AI platform, not found by a chemist screening a compound library. According to Insilico, it's the first time a drug designed this way, target and molecule both from generative AI, has reached a Phase III trial.
Insilico Medicine and SK Biopharmaceuticals announced a $2.5 billion collaboration at BIO 2026 targeting neuroimmune CNS disorders, with just $18 million paid upfront and the rest tied to milestones. The deal structure reveals how pharma is treating AI drug discovery platforms: as high-upside options on unproven pipelines, not proven molecules. - AI drug discovery partnership deal structure - how biotech values AI medicine platforms
The trial itself is unglamorous and specific in exactly the way that makes it checkable: a randomized, double-blind, placebo-controlled study enrolling 320 patients across 47 centers in China, measuring the rate of lung function decline over 52 weeks. China's drug regulator granted rentosertib Breakthrough Therapy Designation back in May 2025, after earlier Phase 2a data was strong enough to publish in Nature Medicine. Insilico closed a $293 million Hong Kong IPO on December 30, 2025, the largest biotech listing on that exchange last year, and rentosertib is the asset most of that money is riding on.
Put the two stories next to each other and you get something more interesting than either alone. One is AI making a diagnostic call in something close to real time. The other is AI generating a molecule years before any trial begins. Different timescales, different failure modes, but the same underlying shift: AI systems are no longer confined to answering questions about medicine. They're inside the actual machinery of it, ordering tests and proposing molecules that then have to survive contact with real patients.
Frankly, the regulatory apparatus is not remotely ready for the first of those two. If an AI agent orders the wrong test or misses a diagnosis inside a live hospital system, nobody has settled who's liable: the hospital, the vendor, or the physician who signed off. The FDA has cleared plenty of narrow AI tools, imaging software that flags a tumor, algorithms that read an EKG, but nothing that autonomously drives a full diagnostic workup the way MIRA did in Dresden's sandbox. Kather's team was careful to say the same thing every serious researcher in this space says: this is not a replacement for doctors, and none of it has been tested on a live patient yet. Insilico's trial, by contrast, sits inside a regulatory pathway that already exists. A drug is a drug, however it was designed, and it still has to clear the same 52-week endpoint everyone else's does.
The gap between those two paths, one with no rulebook yet and one running through the rulebook that's already there, is probably where the real fight over AI in medicine plays out next.
Also read: Bill Gates Says AI Is Powerful Enough to Cause a Billion Deaths • UpGuard found 16000 Supabase databases leaking user data to the open web • The Fed Finally Wrote Stablecoin Rules, and Wall Street Banks Come Out Ahead
This article is posted in AI News, check it out for more related stories.
Metis TechBio's Hong Kong debut shows AI biotech is back in demand Metis TechBio jumped as much as 185% in its May 13 Hong Kong debut after raising about HK$2.1 billion. The move shows investors are paying a scarcity premium for AI-biotech platforms, even when revenue is early and losses remain large. - AI biotech IPO debut Hong Kong stock market - investor demand for artificial intelligence healthcare startups rising
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.