Draft International Treaty on Artificial Intelligence Governance: Model Draft (Version 1.3, complete with Annexes I–VII)

This is a model draft of an international treaty on artificial intelligence governance. It is not a binding instrument and has not been adopted, signed or ratified by any State. The draft is built around human dignity, human agency and the protection of persons in vulnerable situations; effective human control and an obligation of prior constraint (evaluation of an action before it is generated or takes external effect, with three evaluation "gates" for dignity, ethics and necessity/irreversibility, and a duty to hold and escalate to a human); obligations of developers, providers and deployers; transparency, record-keeping and preservation of evidence; independent audit and technical verification; individual rights to know, to human review, to object and to an effective remedy, with protection of complainants and whistleblowers; and international oversight bodies, cooperation and dispute settlement. The full texts of Annexes I to VII (terms, risk classification and objects of evaluation; the Dignity, Ethics and Necessity Gates; hold, human escalation and record-keeping requirements; the prior constraint mechanism; and independent verification and technical verification standards) are included in draft form and remain subject to the drafting reservations; figures and thresholds are left in square brackets for later determination. The final pages list the drafting reservations and matters requiring legal and technical consideration before any diplomatic negotiation. Comments and suggestions are welcome. AI use statement: Artificial intelligence tools were used to assist in drafting, translating and editing this document and its annexes. The author directed the content, reviewed it, and is responsible for the final text.