Signal of the week

HHS released takeaways from its AI request for information and the healthcare sector had three consistent asks

On June 26, 2026, HHS held a public webinar presenting takeaways from the Request for Information it released December 23, 2025, asking how the department can use its regulatory, reimbursement, and research levers to accelerate AI adoption in clinical care. The RFI closed February 23, 2026. Hundreds of responses arrived from providers, researchers, health systems, professional associations, and technology companies.

Ask 1: Cross-agency coordination

Healthcare organizations want a single, coordinated federal AI strategy, not conflicting guidance from FDA, CMS, ONC, NIH, and ARPA-H issued independently. HHS Deputy Chief AI Officer Arman Sharma acknowledged the gap directly: “Too often in government, the right hand doesn't talk to the left hand.”

Ask 2: Implementation and governance support

Healthcare organizations do not primarily need new regulations. They need practical help building AI governance within their specific institutional context. Governance frameworks exist, but as Dr. Deepti Pandita (CMIO, UC Irvine Health) noted at the HIMSS AI Forum: they are not plug-and-play and do not account for differences in resources, culture, or risk tolerance across health systems.

Ask 3: Evaluation and benchmarking tools

Without independent benchmarking standards, procurement decisions default to vendor-supplied performance data, precisely the gap that Mass General Brigham's BRIDGE benchmark (top AI model scored 44.8% on real clinical EHR text despite scoring 92 on medical exams) documented this year. The sector explicitly asked HHS to develop and publish evaluation frameworks.

The clinical evidence check

Mixed structural logic is sound; controlled evidence for governance improving outcomes is early

Do formal AI governance frameworks actually improve healthcare AI safety outcomes?

The direct evidence linking formal AI governance to improved clinical outcomes is limited by the recency of large-scale healthcare AI deployment. Adjacent evidence from aviation, nuclear power, and financial services consistently shows that structured pre-deployment validation, ongoing monitoring, and clear accountability frameworks reduce adverse event rates compared to unstructured approaches.

Within healthcare: a study published in PLOS Digital Health found that hospitals with more formal AI governance practices were significantly more likely to conduct local evaluations of AI tool accuracy and assess for bias across patient subgroups, supporting the argument that governance frameworks drive better validation practices, even before outcome data is available.

No large-scale controlled study has measured whether hospitals with formal AI governance committees have materially better patient outcomes than those without. The practical case for governance is risk mitigation and institutional readiness, not yet a demonstrated clinical outcome benefit.

Regulatory radar

Practitioner voice

On the AI governance challenge: playbooks are not plug-and-play and do not account for differences in resources, culture, or risk tolerance across health systems. The compliance environment is chaotic and layered, spanning federal agencies, state mandates, local hospital policies, and accreditation requirements.

“Bad data will lead to bad AI.” Data governance is the prerequisite for AI governance, not the afterthought. For health systems building AI governance committees: data quality infrastructure must come before the committee structure, not after.

Deal sheet

The Clinical Download  ·  clinicaldownload.com  ·  Manage preferences  ·  Unsubscribe