New places for AI Governance at the UN: UNU-AI and the Independent Scientific Panel

By Javier Surasky



For some time, artificial intelligence (AI) governance has been a topic of international debate. The search for a global governance model capable of combining safety with the continuous development of AI’s potential is the Gordian knot yet to be untied, especially given the implications such a decision—whatever it may be—will have on the race to dominate the AI space, where the three major players have adopted different models: the United States with a market-driven approach, China with centralized state control, and the European Union seeking a balance between risks, citizens’ rights, and technological progress.

In recent months, despite being weakened by its financial situation and its inability to resolve issues such as Russia’s invasion of Ukraine or the unfolding tragedy in Gaza, the United Nations has launched two key initiatives to advance AI governance. It could almost be said that the UN is paving the way for a fourth model, combining elements of the others while at the same time diverging from them. The UN is clearly betting on an AI oriented toward sustainable development, one that factors in the “external” impacts of its growth—such as environmental and social dimensions—addresses the technology gap, and is shaped by experts producing evidence for policymaking.

The first initiative is the creation of the United Nations University Institute on AI (UNU-AI) in Bologna, Italy, as a new academic hub on the subject, guided by the values and principles of the Organization.

UNU-AI is set to become a permanent research institute within UNU, supported by the Government of Italy. Its full operationalization should take place before the end of the year, with a primary mission to mobilize big data and AI for advancing the SDGs, with emphasis on capacity building in the Global South.

The second initiative is the adoption by the UN General Assembly, on 26 August, of resolution A/RES/79/325 on the Terms of Reference and Modalities for the Establishment and Functioning of the Independent International Scientific Panel on Artificial Intelligence and the Global Dialogue on Artificial Intelligence Governance.

This Panel, which will serve as an advisory body to the General Assembly, will be composed of 40 experts from diverse disciplines, nominated by the Secretary-General and appointed by the General Assembly based on expertise, geographical balance, and gender equity. Unfortunately, there are no references to epistemic diversity, nor any acknowledgment of the need to integrate traditional knowledge from Indigenous peoples or direct consideration of vulnerable groups facing AI risks.

Each expert will serve in a personal capacity for three years, with the main objectives of:

  • Producing annual assessments based on independent scientific data on AI’s opportunities, risks, and impacts, modeled after the Intergovernmental Panel on Climate Change (IPCC). Indeed, the design of the Panel is clearly inspired by the IPCC and IPBES (biodiversity protection), as science-policy interfaces delivering authoritative but non-prescriptive evaluations.
  • Maintaining an interactive dialogue with the General Assembly twice a year, and presenting conclusions at the Global Dialogue on AI Governance (GD-AI).

The GD-AI, created under the same resolution, will be a multistakeholder forum meeting annually, alternating between Geneva and New York, to:

  • Facilitate international cooperation and exchange of good practices.
  • Debate ethical, social, cultural, and technical implications of AI.
  • Address technological and human capacity gaps in AI.
  • Promote open-source AI software, data, and models.
  • Reaffirm the primacy of human oversight, transparency, accountability, and human rights in AI development.

The main limitation imposed on the GD-AI is that it shall not address the use of AI for military purposes, leaving this entire field outside its mandate—a political compromise necessary to avoid blocking the whole system.

The greatest risk facing the Panel, and to a lesser extent also the GD-AI, lies in its funding, which will depend largely on voluntary contributions from states, the private sector, and philanthropy, raising doubts about sustainability and the Panel’s true ability to act independently.

Taken together, UNU-AI will focus primarily on the production of expert knowledge, while the Panel will add a dimension of political legitimacy in the multilateral arena, creating opportunities for collaboration between the two.

These initiatives are signs of global efforts to rapidly institutionalize AI governance within the UN system, laying the groundwork for shared norms on AI while taking into account North–South inequalities and current and future impacts, and avoiding normative fragmentation that—given the transnational nature of AI—would lead to ineffective and inefficient results.

Perhaps the most structural challenge facing UNU-AI, the Panel, and the Dialogue alike is their integration into an increasingly crowded landscape of AI forums, such as the OECD’s AI Policy Observatory, the G7’s Hiroshima Process, UNESCO’s initiatives, and the ITU’s AI for Good. Ensuring complementarity and avoiding duplication and competition across these processes, both inside and outside the UN framework, will be critical.