The Emergency Mind Podcast EP 135: Sarah Gebauer on Trust, Failure, and AI Governance
Jun 4, 2026
Sarah Gebauer, a physician who founded Validara Health and leads AI evaluations at MDCalc, discusses clinical AI evaluation and governance. She talks about how hospitals matured AI oversight, translating engineering QA into clinical trust. They explore system-level risks when AI replaces redundancy, AI failure patterns versus human ones, and how interface and role design shape safe use.
AI Snips
Chapters
Transcript
Episode notes
Clinical AI Governance Mirrors Traditional QA
- Clinical AI governance parallels QA and traditional quality processes used in hospitals.
- Sarah Gebauer translated RAND biosecurity model-evaluation methods to healthcare to identify unknown risks and uncertainty around AI failures.
Removing Redundancy Creates Hidden System Risks
- System risks arise when AI removes human redundancy clinicians rely on for safety.
- Sarah warns efficiency-driven removal of humans can eliminate vital 'extra set of eyes' and create new, harder-to-see harms.
Set Governance Before You Deploy AI
- Build governance as the circle around evaluations, covering policy, incident response, privacy, and assigned ownership.
- Ensure local adaptation (e.g., rural hospitals lacking platelets or pediatric ICU) and workforce training before deployment.