Adoption of the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance

On August 26th, the UN General Assembly adopted the institutional designs of the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance. We endorse the final text and we welcome its adoption by consensus. We especially note the role of the co-facilitators from Spain and Costa Rica, whose constructive engagement and persistence were instrumental in achieving this outcome.

 Fig. 1: Structure of the Independent International Scientific Panel on AI in the adopted text. Orange = representative logic, blue = scientific logic.

In this blog, we briefly look back on the negotiations, explain why the adoption of the modalities is an important milestone, and highlight the next steps in turning these institutions into reality.

Areas of contention in the final phase of negotiations

As highlighted in our commentary on Rev. 2 & 3, two points of persistent contention among delegations were the selection of experts for the Panel and the outcome format of the global dialogue. These continued throughout Rev. 4 and the final adopted text.

Expert selection: Member States from developing countries sought to ensure geographic balance in expert selection. Rev. 3 had included a restriction that the main 40-member Panel could have “no more than two selected candidates of the same nationality”. In Rev. 4 this restriction was loosened to “candidates of the same affiliation”. The final text reverted to “candidates of the same nationality or affiliation”.

Even though we support the need for political legitimacy, nationality limits would not have been our preferred option. Still, two factors mitigate the potential negative impact of nationality limits on the speed and politicization of expert selection. First, Panel members serve in personal capacity. Therefore, nationality limits are not positive quotas for politically appointed representatives of large countries. Second, these limits only apply to the 40-member Panel. The Panel can still establish working groups with external experts as needed, with no nationality restrictions. This flexibility is crucial to ensure that the Panel can produce a state-of-the-art synthesis.

Outcome format of the Global Dialogue: Western Member States sought to limit the ambition level for dialogue outcomes. Rev. 3 had foreseen a UN General Assembly resolution that would be adopted prior to the High-Level Review of the Global Digital Compact in the 82nd session of the General Assembly (September 2027 – 2028). In Rev. 4 this was replaced by a declaration open to endorsement by all stakeholders as an input to the High-Level Review. In the final text, the outcome is only specified as an input to the High-Level Review of the Global Digital Compact that considers contributions from different stakeholders.

We also would have been supportive of a higher ambition level for outcome formats. The successful adoption  of parallel US and China-led UNGA resolutions on AI in 2024 indicate that there can still be areas of global agreement on AI. 

The Panel and the Dialogue are an achievement

First, the current geopolitical and financial environment for the multilateral system is challenging. The UN needs to cut staff by 20% across the board. The fact that 190+ UN Member States still managed to find consensus and follow through on designing new institutions should be commended.

Second, the Panel  has the mission of providing a synthesis of the opportunities, risks, and impacts of AI for policymakers, which includes the need for “speed by design”. The global assessment rhythm of one report per year is much more suitable to the domain of AI, than the five to eight year cycles observed in the assessments of the Intergovernmental Panel on Climate Change.

Third, the Panel text reflects a delicate balance between scientific independence and political relevance. The guidance to be policy relevant without being policy prescriptive, combined with the interactive dialogue between the Panel co-chairs and Member States, ensures that the Panel’s reports are not purely academic reviews that end up collecting dust in a drawer. At the same time, guardrails have also been set in place against undue political influence on scientific reports, with scientists serving in individual capacity and having the decision authority on the content of reports.

Fourth, the text explicitly leaves room for the scientists on the Panel to be flexible and agile. This flexibility, ranging from issuing thematic briefs to establishing working groups with external experts as deemed necessary, is not only a crucial element of scientific independence, but also for dealing with the fast-changing nature of AI.

Fifth, the multitrack approach of the Global Dialogue and its organization on the margins of existing UN events with relevance to AI are reasonable. We also echo the need for dialogue on the topics requested by Member States: a) safe, secure and trustworthy AI, b) capacity gaps, c) social and economic implications of AI, d) interoperability, e) human rights, f) human oversight of AI systems in a civilian context, and g) open-source. These topics combine mitigating risks as well as leveraging the benefits of diffusion of AI.

Overall, the modalities negotiations have delivered a functional design of the Panel and the Dialogue. It’s now up to the implementation process to ensure that they live up to their potential.

Implementation timeline

The timeline for the Dialogue is already clear. The Panel timeline is still less defined, but it should present its first annual assessment report during the first Global Dialogue on AI Governance in July 2026. In light of this timeframe, it will be critical for the UN secretariat to advance quickly on the nomination and selection of the Panel.

Dialogue timeline

1. September 2025: High-level multi-stakeholder informal meeting

The Dialogue will be launched in New York during High-Level Week with an informal event organized by the President of the General Assembly.

2. July 2026: First Global Dialogue on AI Governance

The first Global Dialogue takes place on the margins of the AI for Good Global Summit (7-10 July) in Geneva for up to two days.

3. May 2027: Second Global Dialogue on AI Governance

The second Global Dialogue on AI Governance takes place on the margins of the Multi-stakeholder Forum on Science, Technology and Innovation for the Sustainable Development Goals (STI Forum) in New York for up to two days.

Panel timeline

1. Call for nominations

Criteria-based open call launched by the Secretary-General.

2. Selection and appointment

The Secretary-General recommends 40 experts, which are appointed by the General Assembly.

3. Outline & working groups

The Panel creates a high-level outline of what is needed to synthesize existing research related to the opportunities, risks and impacts of AI and creates working groups with external experts for the writing process.

4. Interactive dialogue

The Co-Chairs of the Panel engage in a two-way dialogue with the General Assembly to provide an update on its work and to answer questions.

5. July 2026: First Annual Assessment Report

The Panel presents its first annual assessment report at the Global Dialogue.

Our next steps

We thank all stakeholders for their constructive engagement on the modalities and we look forward to engaging with the implementation process.

As in the institutional design phase, our primary focus in the implementation phase will be on the Panel. Specifically, we hope to help identify and encourage top AI experts to join the Panel, and share ideas on the design of effective working groups.

For any questions or comments, please do not hesitate to reach out to kevin@simoninstitute.ch.