A Global AI Constitution for Health: Ten Rules the World Should Adopt
The thesis
Every jurisdiction is drafting its own health-AI rulebook, and the result will be a Tower of Babel — brilliant models stranded at borders, patients denied tools that are legal one country over. We need shared constitutional principles, not 190 incompatible rulebooks.
The proposed ten articles
- Human accountability is non-transferable. A licensed human is always answerable for a clinical or regulatory decision; AI is decision support, never the decider.
- Right to explanation. Any AI output affecting a person must carry its reasoning and its uncertainty.
- Provenance by default. Every model version, training-data lineage and update is signed and auditable.
- Consent is granular, revocable and portable. Patients own the terms on which their data is used.
- No training on patients without consent. Silence is not consent.
- Prove it before you claim it. No "compliant" or "certified" claim without verifiable evidence.
- Continuous monitoring or no deployment. A model that isn't watched isn't safe.
- Fairness is testable and tested. Publish disparate-impact audits; opacity is disqualifying.
- Least privilege for agents. Autonomy is granted narrowly and revocably.
- Portability of approval. A model validated to a shared standard is presumptively recognised across signatories.
Why now
Standards harmonise fastest before incumbents entrench incompatible ones. The window is open for a few years, then it slams shut.
Provocation
Aviation agreed on shared safety rules so a plane certified once can fly everywhere. Health AI deserves the same — a constitution portable enough to let good models save lives across borders, strict enough that patients never pay for our haste.
A first-principles provocation from the Anxya Health Futures desk. Directional and informational — not medical, legal, financial or regulatory advice. The point is to move the debate, then do the hard validation work.