Independent Consultant on AI & Miltilateral Oragnizational Design

A successful first Dialogue would deliver structural diagnosis rather than aspirational consensus. The current AI governance landscape is not short on principles — it is short on honest accounting of where multilateral mechanisms work and where they don't. Success means acknowledging that global consensus on AI governance is unlikely given deep strategic divergences, and focusing instead on what is practically achievable. Three concrete outcomes would signal credibility. First, positioning the Scientific Panel as a diagnostic tool — producing rapid, timely briefs that identify where institutional capacity is adequate and where it is structurally mismatched to the pace of AI development, rather than comprehensive assessments on multi-year cycles. Second, an honest mapping of where governance approaches genuinely conflict — not just where they nominally overlap — paired with progress on bounded coordination problems: shared risk classifications, common definitions and taxonomies, incident-reporting protocols, and mutual recognition of conformity assessments. Third, establishing a standing mechanism for structured non-state actor input between Dialogues, rather than limiting participation to episodic consultations. The Dialogue should also reflect that AI governance is not only a regulatory or ethical question — it is an operational one. Across the UN system, institutions are increasingly required to make decisions in environments where information cannot be fully verified or attributed. The Dialogue's relevance depends on whether it engages with that reality. The test is not whether this Dialogue produces consensus in July, but whether it builds the capacity to govern at the speed the technology demands — not by forcing coherence, but by enabling translation between frameworks, mutual recognition, and shared protocols. That is a modest ambition, but a credible one.

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

Interoperability is the priority where multilateral action can deliver the most practical value. The EU, the US, and China are pursuing fundamentally different regulatory strategies that are not designed to converge. Most Member States are not shaping these frameworks - they are absorbing downstream effects. The Dialogue's comparative advantage lies not in primary regulation but in enabling translation between frameworks, mapping genuine conflicts, and facilitating mutual recognition where alignment is feasible. Transparency, accountability, and human oversight are selected because AI is lowering the barrier to shaping information environments. What previously required state-level resources is now accessible to a much wider set of actors. This erodes the shared epistemic baseline that institutional decision-making requires. Without meaningful transparency and oversight mechanisms, governance presupposes conditions that no longer reliably hold. Safe, secure and trustworthy AI is included because the operational implications are already present across the UN system. In peace operations, humanitarian response, and development programming, UN entities are often the least technologically capable actors in their operating environments - yet are expected to deliver reliable reporting and exercise sound judgement. Safety and trustworthiness are not abstract principles in these contexts; they are operational necessities. AI capacity-building is selected because the current concentration of AI capabilities among a small number of firms and countries creates structural asymmetries that governance frameworks alone cannot address. But capacity-building cannot only be directed outward. The UN must take its own technological capacity seriously. Across peace operations, agencies, funds, and programmes, the UN is often the least technologically capable actor in its operating environment - expected to verify, report, and exercise judgement while lacking the tools to do so reliably. Without investment in the system's own detection, verification, and analytical capabilities, alongside national and institutional capacity in Member States, the UN risks advocating governance standards it cannot itself meet.

The listed themes do not adequately capture the challenge AI poses to the epistemic foundations of institutional decision-making. AI-enabled manipulation - synthetic media, coordinated information campaigns, machine-speed persuasion - is eroding the shared factual baseline on which governance, diplomacy, and multilateral cooperation depend. This is not a "disinformation" problem to be addressed in a separate workstream. It is a cross-cutting condition that affects every thematic area listed: you cannot ensure accountability without reliable attribution, trustworthiness without verifiable information, or human oversight when the information environment itself has been shaped by automated systems operating faster than human review. The operational dimension of AI governance is also underrepresented. Across the UN system today - in peace operations, agencies, funds, and programmes - institutions are required to make consequential decisions in environments where AI capabilities of other actors exceed what the UN can detect, verify, or counter. This is not a future risk; it is the current operating condition in multiple contexts. The Dialogue risks irrelevance if it addresses AI governance only through the lens of regulation and norms, without engaging with how AI is already reshaping operational decision-making within multilateral institutions themselves. Finally, the listed themes do not address the political economy of AI governance infrastructure. The tools, platforms, and models available to multilateral institutions are overwhelmingly produced by actors with stakes in the very issues being governed. This creates dependency and impartiality challenges that cut across all thematic areas but are not captured by any single one. A cross-cutting workstream on institutional readiness - examining the UN system's own technological dependencies, verification capabilities, and decision-making frameworks - would ground the Dialogue in operational reality rather than regulatory aspiration.

My work spans AI governance, geopolitics, and multilateral institutional design, with a particular focus on peace and security operations. The governance gaps identified above are not theoretical in this sector — they are already shaping operational outcomes. The most significant challenge is the structural mismatch between what multilateral institutions are expected to do and what they can actually verify in an AI-enabled environment. In peace operations, the entity responsible for monitoring a ceasefire or reporting to the Security Council is often the least technologically capable actor in the theatre. Conflict parties, external states, and non-state groups can access AI capabilities — autonomous surveillance, synthetic media, cyber operations — that exceed what UN missions can detect or counter. Decisions must still be made, incidents must still be reported, and briefings must still be delivered, even when the underlying information cannot be reliably established. This is not a future scenario; it is the current operating condition in multiple contexts. The interoperability gap compounds this. Peace operations function across jurisdictions with fundamentally different approaches to AI governance — or none at all. Without shared risk classifications, common taxonomies, or mutual recognition of standards, missions lack even a basic framework for assessing the AI-enabled capabilities and activities they encounter. The opportunity is that the peace and security sector offers a concrete testing ground for governance approaches. The challenges here — attribution under uncertainty, decision-making with unverifiable information, technological asymmetry between institutions and the actors they monitor — are not unique to peacekeeping. They are early manifestations of the governance gaps that will affect every sector as AI capabilities proliferate. If the Dialogue engages with these operational realities, rather than treating AI governance as a purely regulatory or normative exercise, it can produce frameworks grounded in how institutions actually function under pressure.

The Dialogue's most credible role is as a translation and diagnostic mechanism — not as a primary regulator or norm-setter. Global consensus on AI governance is unlikely given the depth of strategic divergences between major powers. The EU, the US, and China are pursuing fundamentally different approaches rooted in different assumptions about the role of the state, the market, and individual rights. These are not converging, and the Dialogue should not pretend otherwise. What the Dialogue can do is make the fragmentation intelligible. That means honestly mapping where governance approaches conflict — not just where they nominally overlap — and identifying bounded coordination problems where alignment is feasible despite broader disagreement. Shared risk classifications, common definitions and taxonomies, incident-reporting protocols, and mutual recognition of conformity assessments are all areas where practical progress is possible without requiring strategic convergence. The Dialogue can also play a role that no other forum currently fills: connecting regulatory discussions to operational reality. AI governance is predominantly framed as a question of standards, principles, and regulation. But across the UN system and beyond, institutions are already making consequential decisions in AI-enabled environments — often without the tools, frameworks, or capacity to do so reliably. The Dialogue should surface these operational governance gaps alongside the regulatory ones. Finally, the Dialogue can serve as an accountability mechanism for the UN system itself. The organization cannot credibly convene global AI governance while remaining one of the least technologically capable institutional actors in the environments where it operates. The Dialogue should include an honest assessment of the system's own readiness — its verification capabilities, technological dependencies, and institutional capacity — rather than directing capacity-building exclusively outward toward Member States.

The Dialogue should build on existing mechanisms rather than duplicating them, and be explicit about where it adds value and where it does not. On standards and technical governance, the work of ISO/IEC, IEEE, and ITU on AI standards is well advanced. The Dialogue's role is not to replicate this but to broker interoperability between these technical frameworks and the political and operational contexts in which they must function — particularly for states that lack the institutional capacity to engage with multiple parallel standard-setting processes. On regional regulatory frameworks, the EU AI Act, China's sector-specific AI regulations, and emerging frameworks in other jurisdictions represent the most concrete governance activity currently underway. The Dialogue should map the substantive tensions between these approaches rather than treating them as complementary building blocks. Honest diagnosis of where frameworks conflict is a precondition for any meaningful interoperability work. On AI and peace and security, the Dialogue should connect with ongoing work within the UN system on AI in the military domain, including relevant Security Council deliberations, as well as operational experience from peace operations, humanitarian response, and development programming. These contexts provide real-world evidence of governance gaps that regulatory discussions alone do not capture. Initiatives like the Secretary-General's Roadmap for Digital Cooperation and the UN system-wide Data Strategy offer institutional precedents for cross-cutting technology governance within the system. On multi-stakeholder engagement, the Dialogue should learn from the limitations of previous processes — including the Global Digital Compact negotiations — where stakeholder input was extensive but episodic, and had limited traceable impact on final outcomes. A standing mechanism for structured input, with transparent feedback on how contributions are reflected in outputs, would represent a genuine institutional improvement over existing practice.

The Dialogue's format should be designed to produce usable outputs, not to maximise the number of interventions. Three structural recommendations: First, move beyond sequential statements. The plenary format of three-minute interventions generates volume but rarely produces the kind of substantive exchange that advances governance. The Dialogue should include working sessions organised around bounded problems — for example, mapping where specific governance frameworks conflict, or identifying minimum viable standards for incident reporting. These sessions should aim for concrete outputs: gap analyses, draft taxonomies, or protocol frameworks that can be refined between Dialogues. Second, differentiate stakeholder roles by function, not category. The current framing — Member States, private sector, civil society, academia, technical community — groups actors by identity rather than by what they can contribute. A more productive structure would organise input around functional needs: who has operational evidence of governance gaps, who has technical expertise on specific coordination problems, who has implementation experience with existing frameworks, and who represents communities affected by governance decisions. This would produce more actionable contributions and reduce the repetition that characterises multi-stakeholder consultations. Third, create accountability for how input is used. A persistent weakness of UN multi-stakeholder processes is that contributions are solicited, acknowledged, and then untraceable in final outputs. The Dialogue should publish a transparent mapping of how stakeholder input informed the Co-Chairs' summary and subsequent preparations. Without this, structured input mechanisms — however well-designed — will lose credibility and participation over successive rounds. On the Scientific Panel specifically, it should establish standing channels for structured evidence submission from practitioners — including from within the UN system — rather than relying solely on academic and technical community input. Operational evidence from the field is essential to grounding the Panel's assessments in institutional reality.

Three categories of underrepresentation deserve attention, and they are not the ones most commonly cited. First, operational practitioners within multilateral institutions. The people making decisions in AI-affected environments — mission leaders, humanitarian coordinators, field-based programme staff — are largely absent from global AI governance discussions. Their experience is treated as implementation detail rather than governance evidence. Yet they are the actors who encounter the gap between governance frameworks and operational reality most directly. The Dialogue should actively solicit their input, including through the Scientific Panel's evidence-gathering processes. Second, states that are norm-takers rather than norm-shapers. The majority of Member States are not developing AI regulatory frameworks — they are navigating the downstream effects of frameworks developed elsewhere. Their governance challenge is not how to regulate AI development but how to adopt, adapt, and maintain agency in an environment shaped by a small number of technology-producing states and firms. The Dialogue's structure should ensure these perspectives inform the substance of discussions, not merely the attendance lists. Third, communities experiencing the consequences of AI deployment without meaningful channels for recourse. This includes populations in conflict-affected contexts where AI-enabled surveillance, autonomous systems, or information manipulation are already present; workers affected by AI-driven labour market shifts; and communities subjected to algorithmic decision-making in public services. Inclusion requires more than invitation — it requires formats that allow affected communities to present evidence on their own terms, and mechanisms that connect that evidence to policy outcomes. What is not needed is another round of broad representation commitments. What is needed is structural design that ensures underrepresented perspectives have traceable influence on outputs — not just presence in the room.

The most effective format innovation would be to stop treating multi-stakeholder consultations as performance and start treating them as working sessions with deliverables. Three concrete proposals: First, structured scenario exercises. Rather than panels followed by statements, the Dialogue should include facilitated exercises that place participants in decision-making roles under realistic conditions. I co-designed and am this week co-facilitating such an exercise for senior peacekeeping leaders — forcing attribution decisions and operational judgement calls in AI-affected environments. The format produces deeper engagement, exposes genuine disagreements, and surfaces governance gaps that abstract discussion does not reach. Adapted for the Dialogue, scenario exercises could confront participants with concrete coordination failures — for example, conflicting risk classifications applied to the same AI system across jurisdictions — and require them to negotiate workable solutions within the session. Second, adversarial mapping sessions. Assign groups to represent different governance approaches — not as advocacy, but as honest articulation of underlying assumptions, interests, and red lines. Then task them with identifying the smallest set of coordination mechanisms that all approaches could accept. This produces more useful outputs than consensus-seeking plenary discussions, because it starts from disagreement rather than pretending it does not exist. Third, rapid evidence reviews. Invite practitioners — from within the UN system, from national regulators, from affected communities — to present five-minute case studies of specific governance gaps they have encountered. No prepared statements; structured evidence followed by facilitated cross-examination. This grounds the Dialogue in operational reality and gives the Scientific Panel primary material that no desk review can produce. The common thread is shifting from input collection to collective problem-solving. The Dialogue has three hours per session. That is enough time to produce concrete analytical outputs — if the format is designed for that purpose rather than for maximum participation.

Several existing approaches offer relevant models - not because they are comprehensive solutions, but because they demonstrate what bounded, practical governance looks like. On regulatory frameworks, the EU AI Act establishes risk-based classification with differentiated compliance obligations. Its value as a reference lies less in its specific provisions than in the structural choice: tiered regulation based on assessed risk. Whether other jurisdictions adopt the same classifications matters less than whether classifications across frameworks can be made mutually intelligible - which is where the Dialogue's interoperability work should focus. On operational experience within the UN system, several entities have built practical understanding of what AI and data governance looks like under institutional and operational constraints. The UN system-wide Data Strategy - which I co-led - demonstrated that cross-cutting technology governance is possible but requires dedicated coordination and sustained political support. Within peace operations, the DPPA-DPO Information Management Unit, the Digital Transformation team, and the Data Strategy team have direct experience navigating technological limitations, data governance challenges, and decision-making under uncertainty. The Secretary-General's Data Action Group has similarly worked across the system on translating data and technology commitments into institutional practice. These actors understand not just what is technically possible but what is institutionally realistic - and where mandates, capacity, and political constraints create gaps that policy frameworks alone cannot close. Their experience should be actively drawn upon by both the Scientific Panel and the Dialogue's preparatory process. On scenario-based governance development, exercises that place decision-makers in realistic AI-affected environments surface governance gaps that policy discussions miss. The exercise I co-designed for the 31st Regional Senior Mission Leadership Course at the International Peace Support Training Centre in Nairobi - confronting attribution and decision-making under uncertainty - produced sharper diagnosis in sixty minutes than months of abstract consultation. This approach could be embedded in the Dialogue's preparatory process.