For most people with idiopathic pulmonary fibrosis, the question is not whether their lungs will fail but how quickly. There is no cure. The available treatments slow the damage; they do not stop it. Median survival after diagnosis is three to five years.
In June, at the BIO International Convention in San Diego, Insilico Medicine announced a collaboration worth up to $2.5 billion in milestone payments with SK Biopharmaceuticals for neuroimmune disorders — a year after publishing Phase IIa trial results for its lead IPF drug in Nature Medicine. The molecule at the centre of this is Rentosertib, a TNIK inhibitor designed to slow the fibrotic damage that kills IPF patients. It is the first drug candidate where both the target and the compound were identified entirely by AI. The platform synthesised and tested 79 candidate molecules; Rentosertib was the 55th. From target identification to clinical candidate: 18 months.
The question is what to do with that.
What 18 months actually means
Traditional drug discovery takes years before a compound reaches a trial. Identifying the right biological target is itself a years-long process — hypothesis, experiment, elimination. Then designing a compound that binds to it, screening variants, testing in the laboratory, a Phase I trial to establish safety. By the time a drug enters efficacy testing, four to six years have typically passed.
Insilico collapsed that phase. The AI generated and assessed virtual molecules computationally, identified the best candidates, and selected one for synthesis without the iterative cycle that makes traditional chemistry slow. The biology remained; the chemistry accelerated.
Rentosertib’s Phase IIa trial enrolled 71 patients across three dose arms, randomised, double-blind, placebo-controlled. Patients on the highest dose showed a mean improvement in lung function of 98.4 millilitres over the study period; the placebo group declined by 20.3 millilitres. For a disease measured in how fast function falls, that gap matters. The results are not definitive; no Phase IIa trial is. But they were published in Nature Medicine in June 2025, and they were positive.
Discovery has changed. The question is whether the rest of the process noticed.
The infrastructure was not built for this
After Phase IIa comes Phase III. Larger trials, typically thousands of patients, with full efficacy and safety data required for regulatory submission. Insilico opened Rentosertib’s own Phase III trial in July, enrolling 320 patients across 47 sites in China. The median time from Phase IIa completion to approval still runs several years, even with expedited review designations. Fast Track status accelerates the conversation with regulators, not the data requirement.
The regulatory requirement is appropriate: evidence from 71 patients is not enough to approve a drug for general use. Phase IIa is promise; Phase III is proof.
But the clinical trial system was not designed around a world in which a promising molecule can be found in 18 months. Recruiting patients, running trials across multiple sites, generating data, submitting for review: the queue was sized for a pipeline that moved at human speed. AI has shortened the front of that process without changing anything that follows it.
For someone diagnosed with IPF today, the arrival of AI drug discovery does not change their timeline. Rentosertib may eventually be approved, or it may not; Phase III trials fail more often than they succeed. The drug was discovered faster, but the path from discovery to patient has not changed. The bottleneck moved from the laboratory to the clinic.
What gets built next
The collaboration announced in June is, among other things, an industry bet that AI will produce candidates at a scale and pace the traditional pipeline never achieved. If that bet is correct — and the discovery-side evidence is accumulating — then the clinical trial system will face pressure it has not faced before: not a shortage of candidates but a surplus of them, each requiring the same years of Phase III data.
Some adaptation is already under way. Adaptive trial designs, accelerated patient matching, synthetic control arms built on existing patient data, earlier integration of real-world evidence from approved therapies — these approaches exist and are being explored. None of them is available to the patient who needs a drug now, and none changes the gap between AI-accelerated discovery and human-speed delivery.
The FDA and the EMA published joint guiding principles for AI in drug development in January, and the MHRA has spent the same period expanding its regulatory sandbox for AI tools. None of it touches the actual constraint. The guidance governs how AI models get validated for use inside a regulatory submission, not how many Phase III trials the system can run at once, or how fast a surplus of candidates gets through it.
My opinion
I think this is the pattern, not the exception. We keep running into the same asymmetry: what a technology like AI can now do at pace and at scale, set against governance and legal procedures that were never built to move at that speed. Some of that caution is earned — Phase III exists because it should be hard to approve a drug on the strength of 71 patients. But most of the controls now standing in the way were built for a more analogue world, and they have not moved into this one. Until they do, every acceleration in discovery just relocates the wait.
For IPF specifically, the constraint is concrete. Median survival after diagnosis is three to five years. The drug that might help was discovered in 18 months. The distance between those two facts is not a scientific problem. It is a system problem — and the system in question is not the one that found the molecule.
You’re reading The Next Evolution by Neil Catton, articles that explore the human world and the intersection of technology, they try and ask difficult questions - not to scare - but to inform. If someone forwarded this to you, you can subscribe free at neilcatton.substack.com.
Neil Catton is the author of The Next Evolution, The Cognitive Crucible and The Shadow System - available on Amazon, and writes at the intersection of technology, ethics, and human purpose.