Artificial Intelligence Needs an International Agency – Now

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Artificial intelligence is advancing faster than the international institutions responsible for addressing its cross-border risks. Its benefits are substantial, but increasingly capable systems also create externalities, information asymmetries and competitive pressures that no individual company or national authority can address alone. The mechanisms established by the United Nations in 2025 provide scientific assessment and international dialogue, but they lack a mandate for mandatory declarations, independent evaluation, compliance verification, incident investigation and coordinated response. Recent incidents and public statements by researchers and industry leaders sharpen the urgency of this institutional gap, but the case for action does not depend on any single event, estimate or forecast. This article therefore recommends that the United Nations General Assembly formally open a pathway toward an International Artificial Intelligence Agency. Governed by its own statute, the agency would have a dual mandate for promotion and verification, with enhanced scrutiny limited to frontier AI activities exceeding internationally agreed thresholds for risk, autonomy, reach or cross-border impact. Its functions would include technical cooperation, capacity-building, declarations, international standards, accreditation of independent evaluators, incident reporting and investigation, verification of agreed controls and formal findings of non-compliance. The agency would conduct independent technical fact-finding, while political consequences would remain with States and the competent United Nations bodies. The proposal does not assume that AI can be verified in the same manner as nuclear material or that any evaluation can certify absolute safety. Verification would instead combine proportionate methods to assess compliance with agreed obligations and create an independent factual record, while protecting confidential and proprietary information. The immediate request is not to approve a final verification methodology or create a fully operational agency overnight. It is to authorize an institutional process that can begin testing methods, building international capacity and preparing a negotiated framework before a major crisis forces an improvised response.