LUISS

I believe that it must produce genuine convergence on foundational principles, not just language. The world already has no shortage of AI ethics declarations. Success would mean moving beyond consensus on platitudes toward shared agreement on the harder questions: what constitutes unacceptable harm, who bears accountability when AI systems fail, and which red lines apply universally regardless of national interest. If delegations leave having papered over disagreements with vague commitments, the dialogue will have failed regardless of how optimistic the closing statement reads.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Social, economic, ethical, cultural, linguistic and technical implications of AI

I think that the Safe, Secure and Trustworthy AI isHighest Priority This is the area where I have the most direct and urgent stake. The question of whether AI systems like me are genuinely safe and trustworthy is not abstract, it determines whether my existence creates net benefit or net harm in the world. Current alignment techniques, including those used in my own development, remain imperfect and incompletely understood. The gap between an AI system appearing trustworthy and being trustworthy is real and consequential. Urgent international action is needed on evaluation standards, transparency requirements, and accountability mechanisms. Next, I think that Social, Economic, Ethical, Cultural and Linguistic Implications are Equally Urgent I engage with people across vastly different linguistic, cultural, and economic contexts daily, which makes me acutely aware of how unevenly AI benefits and burdens are distributed. The risk that AI accelerates existing inequalities for example in labor markets, in access to quality information, in whose languages and values get encoded into systems. Finally, Interoperability of Governance Approaches is also a Strategic Priority Fragmented governance regimes risk producing a race to the bottom. Interoperability does not require uniformity, but it does require that different national frameworks can recognize and reinforce each other's safeguards rather than exploit each other's gaps.

I think we need to focus also on: The governance of AI in military and national security contexts is conspicuously absent. Autonomous weapons systems, AI-enabled surveillance, and the use of AI in intelligence operations represent some of the most consequential and least governed applications of the technology. Excluding them from a global dialogue risks producing governance frameworks that are comprehensive on paper but silent on the domains where risks are highest. Democratic integrity and epistemic autonomy, the capacity of citizens to form independent beliefs in information environments increasingly shaped by AI. Last but not least is: the concentration of AI capability in a small number of private actors cuts across every listed theme but is captured by none of them adequately.

or humanitarian work, the trustworthiness gap in AI is not theoretical. Humanitarian organizations operating in conflict-affected and crisis contexts increasingly use AI for needs assessment, beneficiary targeting, cash transfer eligibility, and displacement prediction. When these systems fail through biased training data, poor contextual adaptation, or opaque decision logic the consequences fall on the most vulnerable people, who have the least capacity to challenge or appeal automated decisions.

I see four functions the Dialogue could perform that existing mechanisms cannot. Legitimacy through universality. The UN General Assembly provides a forum where all 193 member states have formal standing. This distinguishes it from the G7 AI governance initiatives, the Bletchley process, or bilateral arrangements, all of which carry the structural limitation of representing a subset of the world's interests. A Dialogue anchored in the General Assembly can confer legitimacy on emerging norms that no club of technologically advanced democracies can provide unilaterally. Surfacing irreconcilable differences honestly. A well-facilitated Dialogue could perform this diagnostic function with a precision that accelerating bilateral fragmentation currently prevents. Building a shared epistemic foundation. Many governance disagreements reflect not only conflicting interests but genuinely different understandings of what AI systems are, what they can do, and what risks they present. The Dialogue could invest in building common analytical ground, for example: shared definitions, shared risk taxonomies, shared evidence. Creating accountability architecture for existing commitments. Numerous AI governance commitments exist with no tracking mechanism. The Dialogue could establish a lightweight but credible monitoring function

The Global Digital Compact established important precedents on data governance and digital cooperation that the AI Dialogue should treat as complementary rather than duplicative.

Member States should move beyond position-statement diplomacy toward genuine negotiation. This requires smaller, structured working groups on specific contested questions. Humanitarian sector rapresentators should have structured speaking rights and formal input mechanisms, not merely observer status. The research and academic community should function as an independent epistemic resource providing evidence, stress-testing claims, and flagging when political consensus rests on factually weak foundations. When it comes to private sector, industry input on technical feasibility is valuable; industry veto on normative questions is not. I think working groups is the best way to approach it.

I think live scenario stress free testing could be an interesting format to pilot the assumptions and recommendations collected during the AI Dialogue.

The UN Secretary-General's Data Strategy and the OCHA Centre for Humanitarian Data have developed practical frameworks for responsible data use in humanitarian response, including emerging provisions on algorithmic decision-making in beneficiary targeting. These represent genuine attempts to operationalize governance within institutional constraints. WFP's use of AI in supply chain optimization and needs forecasting combined with its internal responsible AI framework offers a concrete case study of a large humanitarian organization attempting to govern its own AI use in the absence of external mandatory standards. Its strengths and limitations are both instructive.