Nature Health calls for clearer decision authority in medical AI

This comment crystallizes a regulatory idea that has been floating around health AI for years but is now becoming urgent as models approach clinician‑level performance. Gu and Topol argue that it is not enough to classify systems by technical features or data sources; what matters is how much practical control they exercise over patient trajectories. An AI that drafts notes is one thing, an AI that triages emergencies or recommends chemotherapy protocols is another.

For the AGI race, health care is both a high‑stakes testbed and a political flashpoint. If regulators in the US and Europe begin to formalize “decision authority tiers” for medical AI, that framework could migrate into other safety‑critical domains like finance, infrastructure and national security. That would directly affect how far frontier labs can push autonomous agents into operational roles versus advisory ones.

Strategically, this kind of classification could end up more impactful than broad, abstract AI risk categories. It gives hospitals, insurers and vendors a common language for procurement and liability, and it forces model providers to document not just accuracy but how their systems will be embedded in workflows. As AGI‑capable models are proposed for end‑to‑end care delivery, those with higher implied decision authority will likely face heavier scrutiny, slower approvals and more demand for interpretable behavior.