HHS launches ARPA-H SURPASS to speed clinical trials with models and agents

The Department of Health and Human Services, through ARPA-H, said Wednesday it launched SURPASS — Simulation-augmented, Real-time Platform Adaptive Seamless Trials — plus companion projects STACK, COMMONS, and CINCH to redesign how drugs and biologics are tested in the U.S.

A new drug still often takes more than ten years and up to $2 billion to test, and most of those efforts fail. SURPASS is the federal bet that computer models of patients, statistics that stay valid while a trial is running, and software that handles the startup work can shorten that path without dropping the standard of evidence. It is not a new medicine. It is a new way to test whether a medicine works.

On Wednesday, 30 September 2026, the U.S. Department of Health and Human Services, through the Advanced Research Projects Agency for Health, said it launched SURPASS. HHS is the federal health department. ARPA-H is the research agency inside it. The record is the ARPA-H news page, “HHS launches SURPASS and new efforts to accelerate faster, smarter clinical trials,” and the same announcement in the HHS press room. Both pages print September 30, 2026. Neither page prints an hour. The dateline is Austin, Texas. SURPASS stands for Simulation-augmented, Real-time Platform Adaptive Seamless Trials. The pages say the department launched an effort to change how clinical trials are designed and run, using computer models, real-time analysis, shared infrastructure, and automation. The aim, as the pages state it, is to evaluate drugs and biologics faster, at lower cost, and with fewer participants. A clinical trial is the study that tests a treatment in people. A biologic is a medicine made from a living source, such as a protein or an antibody, rather than a chemical mixed in a lab. The same announcement names three companion projects: STACK, COMMONS, and CINCH. Those lines are HHS and ARPA-H.

The problem, in the agency’s figures. The news pages say clinical development of drugs and biologics often takes more than a decade, costs up to $2 billion on average, and fails more than 90 percent of the time. A decade is ten years. Up to $2 billion on average is the agency’s cost line. It is not a bill for one named drug. More than 90 percent means more than nine in ten of those development efforts fail. The pages say promising treatments never reach patients, either because they cost too much and take too long to clear a regulator, or because they will not earn enough once they are approved. The same pages say the United States risks losing the lead in clinical trials, and the economic activity that comes with that work. The SURPASS program page prints a different pair of figures for the same problem: $1 billion to $2 billion, written on the page as “$1-$2billion,” and a failure rate of 90 percent, not more than 90 percent. Both pairs are ARPA-H’s. They are not one number restated.

What the program is supposed to build. The news pages say SURPASS seeks continuous, adaptive platform trials. Those trials would combine predictive computer models, shared trial infrastructure, common control groups, and real-time analysis, so a team can decide earlier whether a drug is working. The program page uses perpetual, where the news pages use continuous, for that same kind of platform trial. A platform trial is one study setup that can test more than one treatment, instead of starting over for each drug. Adaptive means the trial can change, under its own rules, as results come in. A control group is the comparison group, the people who do not receive the new treatment. A common control group is one comparison group shared across treatments, instead of a new one for every drug. The news pages say more efficient designs would let more patients join trials that are more likely to help them. The program page says that if the work succeeds, patients would get promising treatments sooner, because trials would take less time, cost less, and ask less of the people in them. If successful is the page’s condition. It is not a finished result.

Three technical areas, in the order the news pages print them. The first is a phaseless design engine, for faster trials with fewer patients. Ordinary drug testing stops and restarts in numbered phases. Phaseless, on this page, means a design that does not keep restarting between those phases. The engine would bring digital twins and other predictive models into the design, simulate how the trial would go before it starts, and build the evidence a regulator would need before trusting those methods. A digital twin, here, is a computer stand-in for a patient or a trial, used to try an outcome before it happens in people. The second area is a continuous inference engine, for faster decisions. Inference means drawing a conclusion from the data. The page says the engine would support analysis that stays valid in real time, or whenever a team asks for it, allow the trial to adapt quickly, and shrink the need for a large conventional control group, while still producing rigorous evidence. Always-valid means the statistics still hold if the team looks while the trial is running, not only at one planned look. The third area is an agentic operations layer, to cut time and cost. Agentic, here, means software that takes a next step, such as starting a trial site or cleaning a dataset, rather than only answering a question. The page says that layer would automate trial startup and day-to-day operations, help add a new treatment arm, and speed data collection, cleaning, and the building of the dataset. A treatment arm is one group in the trial, such as the people receiving a particular drug. Those three lines are HHS and ARPA-H. The announcement does not say any of the three is built.

Three companion projects, announced the same day, for problems outside the trial design itself. STACK is meant to add clinical sites in the United States and make enrollment more efficient. The project would use artificial intelligence to speed site activation and turn research-naïve sites into places that can run a study. Research-naïve means a clinic that has not run trials before. The page’s aim is more capacity, and more patients able to join research closer to home. COMMONS is the data layer. It would build privacy into the design of consent at a national scale, and open access to regulatory-grade data. Consent is a person’s permission to use their information. Regulatory-grade means careful enough that a regulator could rely on it. The page says that data would matter for fast enrollment, for the predictive models, and for the rest of the modernization work. CINCH is about the patient. It would help people contribute their own real-world data and benefit from it: care that continues, better coordination, and a faster match to a trial that might fit. Real-world data is health information from ordinary care, not only from a trial visit. The pages say the four efforts together take on outdated designs, too few sites, fragmented data, and a patient experience that is hard to navigate. Those lines are HHS and ARPA-H.

Who is quoted, and whose words they are. HHS Secretary Robert F. Kennedy, Jr. said America should lead the world in turning medical breakthroughs into treatments. He said HHS is taking on the delays, the duplication, and the unnecessary costs that slow trials down. He said that through ARPA-H, the department is building a modern clinical trial system that can generate rigorous evidence faster, reduce the burden on patients, and move effective treatments from discovery to the people who need them sooner. That quotation is his, in the release. SURPASS program manager Daria Fedyukina, Ph.D., said today’s system often requires too many stops, too much duplicated infrastructure, and too much time before researchers know whether a trial is on the right track. She said SURPASS is developing the technology and the statistical methods to analyze trial data as it comes in, so teams can learn in real time, adapt when they need to, and produce stronger evidence faster, with less operational burden. The release says she also oversees STACK and COMMONS. She said trials are slowed by site activation, broken-up data access, and an experience that can be very hard on patients and caregivers. She said STACK and COMMONS are meant to take those barriers on, so more patients can reach a trial and more trials can run in the United States. Program manager Erika Kim, Ph.D., oversees CINCH. She said patients do not only take part in trials. Their real-world data and what they live through are needed to power the research. She said CINCH will make it easier to contribute that data, so the research reflects patient experience, supports care that continues, and connects people faster to a trial that may fit. Those quotations are in the release. A quotation is not a patient already enrolled.

What the announcement is, and what it does not say. The news pages describe a program launch and three companion projects. They do not name a company or a university that has been paid to do SURPASS. They do not print a dollar award for the program. They do not say the Food and Drug Administration has approved a drug, a model, or a trial design under SURPASS. The HHS page points readers to the ARPA-H program page for an Innovative Solutions Opening, the formal call for proposals, and to an Award Directory for COMMONS, CINCH, and STACK. The program page says the funding opportunity is open. The solicitation notice id is ARPA-H-SOL-26-164. The full documents are posted on SAM.gov, the federal site for government contract notices. The overview lists four dates. An informational webinar is 15 October 2026. Registration for it closes 13 October. A proposers’ day is 6 November 2026. Registration for that day closes 31 October. A solution summary is due 30 November 2026, and the page says a summary is required before a team can continue. A full proposal is due 22 January 2027. Those dates are the schedule for teams that want to bid. They are not a date when a drug reaches a patient. The news pages also say the work advances the Trump administration’s effort to modernize clinical development, cut delay and cost that are not needed, and strengthen U.S. standing in biotechnology and drug development. That sentence is the release’s.

The picture is the official ARPA-H social card for SURPASS. On a blue and purple field, SURPASS is set in large white type. Under it, the card reads “Simulation-augmented, real-time Platform Adaptive Seamless Trials.” An abstract graphic of cyan and violet bars sits to the right. The ARPA-H mark, a white H in a hexagon, is in the corner. The card names the program and its full title. It does not print a calendar date.

In plain terms, HHS said on Wednesday that ARPA-H opened SURPASS, a program to change how the country tests drugs and biologics. The agency’s news pages put the problem at more than ten years, up to $2 billion on average, and a failure rate above 90 percent. The tools it names are digital twins inside a trial design that does not keep restarting, statistics a team can read while the trial is still running, and software that handles startup and data work. STACK, COMMONS, and CINCH are the same-day projects for trial sites, shared data, and patients. The program page shows an open call for proposals, with a summary due 30 November 2026 and a proposal due 22 January 2027. The announcement does not name a winner, and it does not print a dollar award.