By Steve Herne, CEO of Unlearn

Every clinical trial is a long chain of decisions, and the strongest programs are the ones that get them right. Looking back at 2026 and heading into Q4, that's the perspective driving our work at Unlearn: helping sponsors make the best possible decision at every point in a trial with digital twins, data, and AI.

The clearest evidence of that was in our science. In the space of about two weeks this spring, our scientific team put out three new papers, and what connects them is the thing they each make more reliable: the estimate of whether a drug works. One paper shows how digital twins can serve as synthetic controls for single-arm trials without a control group, with worked examples in ALS and Huntington's disease. 

The second paper, which we call FRESH, takes on a problem that arises whenever the population you actually want to study has no clean, matched dataset available. The evidence usually exists. It has been published, reviewed, and written into a summary table that cannot be compared against patient-level data. FRESH is a way to bring what sits in those tables into a patient-level model, so a team can reason about a population for which nobody ever handed them patient-level data. We first put it to work on an oncology question where the head-to-head trial a sponsor would want has never been run. FRESH is also now being applied in obesity and neuropsychiatric disorders.

And the third paper takes on the regulatory question directly: how a prognostic method like PROCOVA can prospectively shrink the sample size of a registrational trial under the FDA's draft guidance on AI, illustrated in Alzheimer's. We wrote that one with Tala Fakhouri, who authored the FDA's first AI guidance and is now Chief AI and Regulatory Strategy Officer at Parexel. All three are available as preprints; we publish this way on purpose so that the people best equipped to challenge the methods can do exactly that.

Those regulatory questions deserve more than a preprint, which is part of why we're taking them up live. On September 28, we're hosting a webinar, More Drugs, Faster: Can Clinical Development Keep Up?, bringing Tala together with Barbara Bierer, faculty director of the Multi-Regional Clinical Trials (MRCT) Center of Brigham and Women's Hospital and Harvard, and our co-founder Jon Walsh, for a frank conversation about what the FDA's evolving position on AI means for the teams designing and running trials today. If that's a live question for you, I hope you'll join us.

You can see the same aim in the partnerships we announced. In July, we began a collaboration with Acumen Pharmaceuticals to conduct analyses of their Alzheimer's disease programs, which are among the hardest places to run efficient trials. In ALS, our work with VectorY Therapeutics supports their PIONEER-ALS study, a Phase 1/2 trial of a first-in-class vectorized antibody therapy, where an individual digital twin for each participant lets the team draw a clearer read from every patient enrolled. Two diseases, two very different studies, and the same question underneath each: how do we help this team decide well with the patients and the data they actually have?

Oncology, still our newest space, was a bright spot. Our team held a record number of meetings at ASCO, the world's largest oncology meeting, and the FRESH work gave those conversations something concrete to stand on. We stayed active on the scientific circuit throughout, from AACR and ASCO in the spring to a poster at AAIC in London and an invited talk at AACR's Drug Discovery and Development meeting over the summer. These are the rooms where our work is tested by the people best equipped to do so, which is exactly where it should be.

What's become clearer to us this year is that the constraint FRESH removes was never really an oncology constraint. Sparse data is the norm across much of drug development: rare diseases, new mechanisms, populations defined narrowly enough that no one has ever assembled a cohort. One of the reviewers on the FRESH paper made the point to us better than we had ourselves: that nothing about the approach is inherently limited to oncology. We think that's right: we've already begun applying it in obesity and neuropsychiatric disorders, and where we point it next is one of the more interesting questions for our science team.

In the middle of all this, a signal from outside the company came that I don't take for granted. Unlearn was named the winner of the 2026 Fierce AI Innovation Award for Clinical Trial Design, in the award's inaugural year. What makes that recognition meaningful to me is its basis: real-world impact, judged by people with a full view of the field, at a moment when there is no shortage of AI hype to cut through. Being recognized for work that is scientifically rigorous, clinically meaningful, and trusted by both customers and regulators is exactly the standard we want to be held to.

None of this is finished, and that's the point. A trial is a chain of decisions, each one shaping the next, and our job is to help make each one a little more grounded than it would otherwise be. If you're designing a study, in the middle of one, or trying to make sense of a readout, that's a conversation we'd welcome.

Q4 Conferences

If you'll be at any of these, come find us — we'd love to hear more about your clinical program.

• European Society for Medical Oncology (ESMO) Congress — October 23–27, Madrid

• Northeast ALS Consortium (NEALS) — October 27–30, Clearwater, FL

• Society for Immunotherapy of Cancer (SITC) — November 4–8, Phoenix

• American Medical Informatics Association (AMIA) Annual Symposium — November 7–11, Dallas

• Obesity Week — November 14-17,  Washington DC

• Clinical Trials on Alzheimer's Disease (CTAD) — November 16–19, Boston

• International Symposium on ALS/MND — December 9–11, Amsterdam