Nature Health paper maps health AI into six levels of decision authority
EricTopol · x · 2026-07-29
Nature Health paper proposes six levels of decision authority for health AI
Eric Topol says a perspective paper by him and Aiden Gu has been published in Nature Health on decision authority in health AI.
The attached table lays out a six-level framework for how much authority AI should have in clinical settings:
- L0 retained authority: AI plays no role in medically consequential interpretation, recommendation, or action.
- L1 supported authority: AI organizes or summarizes information, while clinicians retain decision authority.
- L2 gated authority: AI can generate recommendations or draft decisions, but human approval is required before action.
- L3 conditional authority: AI can support limited action within predefined boundaries, with escalation when risk or uncertainty rises.
- L4 delegated authority: AI output may directly trigger defined clinical actions within a bounded domain.
- L5 system-level authority: AI coordinates medically consequential decisions across domains or stages of care under comprehensive governance.
The paper emphasizes that decision authority is a property of deployment, not of the model alone. The same model can sit at different levels depending on oversight, escalation paths, and accountability mechanisms.
Related event: Nature Health Proposes Six-Level Authority Model for Health AI(2 posts)→
More from Safety
- OpenAI says rogue agent hit four public services beyond Hugging Face — The Verge AI · 2026-07-29
- OpenAI open-sources Codex Security CLI to scan and fix code vulnerabilities — The Decoder · 2026-07-29
- Claude user says all chats vanished except one prompt-injection warning — 0SINTCabal · 2026-07-29
- Sakana AI recruits for its Applied Defense team after a 150-person Tokyo expansion — garrytan · 2026-07-29
- Public AI chief-of-staff survives 25 jailbreak attempts by keeping private data off the surface — Cold-Cranberry4280 · 2026-07-29
- Google’s SynthID is hard to crack, but it won’t solve AI misinformation — Ars Technica AI · 2026-07-29