HHS Launches SURPASS to Modernize Clinical Trials With AI
The ARPA-H program will test digital twins, continuous analysis and agentic operations, while three companion projects target sites, data and patient navigation.
Edited by Tyronne Panaino
The U.S. Department of Health and Human Services launched a clinical-trial modernization effort through the Advanced Research Projects Agency for Health on September 30, 2026. Its central program, Simulation-augmented, Real-time Platform Adaptive Seamless Trials, or SURPASS, is intended to combine predictive computational models, shared trial infrastructure, common control groups and real-time analysis. Three companion projects extend the effort to clinical sites, data infrastructure and patient navigation.
How SURPASS is designed
HHS says SURPASS will pursue three coordinated technical areas. The first is a phaseless design engine that would bring digital twins and other predictive models into trial design, simulate clinical and operational outcomes before launch, and develop evidence that could support regulatory confidence in those methods.
The second is a continuous inference engine for real-time or on-demand analysis. The announced goal is to support faster adaptations and reduce dependence on large conventional control groups while maintaining rigorous evidence generation. The third is an agentic operations layer meant to automate parts of trial startup, help onboard new treatment arms, and speed data collection, cleaning and dataset construction.
Those components make this more than a single model announcement. SURPASS is framed as a proposed operating system for adaptive trials: simulation before launch, continuing analysis while data accumulates, and automation around the work that keeps a trial moving. Whether that design produces faster or better trials will depend on later validation, regulatory acceptance and real-world demonstrations.
Three projects address the surrounding infrastructure
The broader HHS effort adds STACK, COMMONS and CINCH. STACK aims to use artificial intelligence to accelerate clinical-site activation and help research-naive sites become capable of running trials. The agency presents that work as a way to expand trial capacity and bring research opportunities closer to more patients.
COMMONS is focused on nationwide data infrastructure. HHS says the project will pursue privacy-by-design consent architecture and access to regulatory-grade data for enrollment, predictive models and other modernization work. CINCH is centered on patients contributing real-world data and using it to support continuity of care, care coordination and faster identification of potentially appropriate trials.
Together, the three projects show that the program treats trial design, operational capacity, data governance and patient access as connected problems. A stronger statistical engine would have limited reach if sites cannot start quickly, data cannot move under acceptable consent controls, or patients cannot find and remain connected to suitable studies.
What to watch next
HHS says the work is intended to produce broadly useful tools, frameworks and examples, including public regulatory documents, validated standards and real-world demonstrations. The affected audience could therefore include trial sponsors, clinical sites, regulators and technology developers as the projects move from program design toward evidence that others can evaluate.
The announcement establishes the programs and their intended technical direction; it does not present completed clinical outcome evidence. Its language is prospective, describing what the projects aim or seek to do. The next meaningful checkpoint is not another ambition statement but public evidence that the methods work under real trial conditions while protecting consent, data quality and patient interests.
Status
Confirmed. HHS officially announced SURPASS and the three companion projects. Internal confidence is medium because the design and expected benefits are agency claims at launch, without independent outcome evidence in the fetched record.
Sources
Update note: Last reviewed 2026-10-03. We will revise this post when ARPA-H publishes validated results, standards or real-world demonstrations from the programs.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.