Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the AI In Healthcare topic
No spam. Unsubscribe anytime.
Panelists urge risk‑based oversight for AI while pressing for practical privacy solutions in behavioral health
Summary
Industry and federal representatives discussed AI uses (ambient scribing, summarization, predictive models), implementation caveats, clinician workload risks, and persistent privacy and consent challenges (42 CFR Part 2), urging risk‑based oversight and more evidence across care settings.
Industry representatives and clinicians at the ONC roundtable described multiple, practical AI uses in behavioral health — from ambient scribing and summarization to agentic services that surface the most relevant information at point of care — while warning that real‑world benefits depend on implementation, clinician workflow integration, and evidence of time savings.
Dr. Rowland Illing of Amazon Web Services highlighted infrastructure and ‘‘agentic services’’ that can execute back‑office tasks and enable ambient voice capture and automated code extraction. Tim Platts of Epic described how AI can summarize large volumes of data, surface risk factors and support predictive models for suicide risk and readmission. He said ambient tools let clinicians focus on patients, noting that telehealth usage remained high for mental health since the COVID surge.
Clinicians and policy leads balanced enthusiasm with caution. Dr. Michelle Durham said some early studies show modest time savings from AI scribes and cautioned that rushed or poorly integrated implementations could add to clinician burden. John Snyder (HRSA) and others warned that patients and providers have privacy concerns about ambient listening and the need to harmonize federal and state rules. Tim Platts and several clinicians singled out 42 CFR Part 2 — the federal rule protecting certain substance use disorder records — as an area where differing interpretations and conservative implementations can create dangerous information gaps.
Panelists recommended pragmatic next steps: pilot deployments with risk‑based oversight, expanded evaluation of AI tools across FQHCs and CCBHCs as well as academic centers, clear consent workflows, and targeted funding for implementation so clinicians receive both the technology and the training needed to use it safely and effectively.
The discussion closed with broad support for oversight calibrated to risk and functional pilots to build the evidence base before large‑scale mandates.
