On October 6, ONC announced two LEAP in Health IT awards totaling $1.5 million. The agentic-AI project, led by MedStar Health Research Institute, will build open-source middleware for long-running clinical follow-up using patient-gated data pods, verifiable AI routing and tamper-evident certificates. This is research funding—not a new provider mandate.

DomAfina’s take: home care software should borrow the safety pattern before borrowing the autonomy. Any AI that schedules a visit, routes a task or follows up with a patient needs a named human owner, narrow permission scope, a patient-consent record, a visible stop control and a timestamped log of every input, action and handoff. Test failure modes with synthetic patients, then require human approval for anything that changes care, billing or medication workflows. An agent that can act must also prove who allowed it, what it did and how to stop it.

Source reviewed: ONC: HHS Announces 2026 LEAP in Health IT Awardees. The source supports the reported update; the operational recommendations are DomAfina’s analysis.