Academy for AI Diplomacy and International Affairs (in formation)

A successful first Global Dialogue on AI Governance would produce three concrete outcomes. First, a clear recognition that AI governance must extend beyond economic and social domains to include the protection of civilian digital infrastructure – specifically data centers, undersea cables, and energy grids – from AI-enabled targeting. Existing international law protects hospitals and museums but remains silent on these civilian assets. Second, a commitment to capacity building for the 89 countries that currently cannot convene three people to discuss AI governance. The UN Secretary-General's High-Level Advisory Body on AI has identified this gap. The Global Dialogue should produce a concrete action plan, not just statements. Third, an open-source repository of technical safeguards – zero-knowledge verification protocols, API kill switches, cryptographic watermarking – that any State, humanitarian agency, or private sector actor can adopt immediately, regardless of whether formal treaties are in place. The Global Dialogue does not need to produce a binding treaty. But it should identify model frameworks that coalitions of willing States can adopt, just as the Ottawa Treaty on landmines emerged from a coalition outside the UN consensus process.

• Safe, secure and trustworthy AI
• AI capacity-building
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

I selected these four priorities because they are the areas where concrete, actionable progress is possible without waiting for formal treaty ratification. Safe, secure and trustworthy AI: The Digital Hague Convention (2026) defines when AI targeting systems become autonomous weapons (≥80% human approval of AI recommendations with <30 seconds per review). This provides an operational definition of "meaningful human judgment" - a gap in existing international humanitarian law. AI capacity-building: I authored a field manual, "How to Train 100 Diplomats on AI for $0," using open-source models (DeepSeek, Mistral). A diplomat anywhere can produce an intelligence briefing in 7 days. The UN can distribute this manual at no cost. Transparency, accountability, and human oversight: The treaty requires explainable AI for targeting systems - no black boxes. It also establishes corporate criminal liability for technology companies that knowingly enable prohibited targeting, and universal jurisdiction for prosecution. Open-source software, open data and open AI models: The treaty's technical annexes (zero-knowledge verification, API kill switch, cryptographic watermarking) are open source. Any UN agency - UNHCR, WFP, ICRC - can adopt them for humanitarian data protection immediately. These priorities are interdependent. You cannot have safe AI without transparency. You cannot have capacity building without open-source tools. And you cannot have accountability without legal frameworks.

Yes. Two cross-cutting issues are missing. First, the protection of civilian digital infrastructure in conflict. The listed themes focus on safe, secure, trustworthy AI - but do not explicitly address what happens when AI systems are used to target civilian infrastructure (data centers, undersea cables, energy grids) in armed conflict. This is not a niche issue. Documented incidents include the severing of undersea cables in the Baltic Sea and Red Sea, AI-generated kill lists with 20-second human reviews, and the weaponization of everyday communication devices. The Digital Hague Convention (2026) fills this gap. Second, the intersection of AI governance and international humanitarian law. The Global Dialogue's mandate is civilian AI governance. That is appropriate. But AI systems developed for civilian purposes are being repurposed for military targeting. There is no international legal framework governing this transition. The Global Dialogue should acknowledge this gap and refer it to relevant UN bodies (e.g., the CCW process on lethal autonomous weapons systems) for further work. I am not suggesting the Global Dialogue expand its mandate into armed conflict. But a one-sentence acknowledgment - "AI governance in civilian domains has implications for international humanitarian law that require further examination" - would provide political cover for other UN processes to act.

As a civil society actor based in Geneva, working at the intersection of AI governance and international law, I observe two significant governance gaps affecting the digital rights and humanitarian sectors. Challenge: Legal ambiguity for civilian digital infrastructure. Current international law does not explicitly protect data centers, undersea cables, or energy grids from AI-enabled targeting during armed conflict. Some States treat data as a protectable object; others deny it any status. This uncertainty directly harms humanitarian operations. In 2022, the ICRC was forced to suspend family reunification services for over 500,000 people after servers hosting refugee data were breached. Belligerents exploit legal ambiguity. Humanitarians bear the cost. Opportunity: Open-source technical safeguards exist today. The governance gap is not technological – it is legal and political. Zero-knowledge verification protocols, API kill switches, and cryptographic watermarking are all technically mature. The Digital Hague Convention (2026) packages these as open-source annexes. Any UN agency – UNHCR, WFP, ICRC – could adopt them immediately, regardless of treaty ratification. This is an opportunity for the Global Dialogue to promote open-source, implementation-ready tools alongside principles and declarations. Challenge: Capacity asymmetry. The UN Secretary-General's High-Level Advisory Body on AI identified 89 countries that cannot convene three people to discuss AI governance. This capacity gap is not being closed by existing initiatives. My sector (civil society AI governance) lacks a zero-cost, low-infrastructure training pathway for diplomats. I have drafted a manual to fill this gap, but distribution remains a challenge. Opportunity: Coalitions of the willing. Like the Ottawa Treaty on landmines, AI governance does not require universal consensus to advance. The Global Dialogue can identify model frameworks that coalitions of States can adopt, adapt, and cite as precedents.

The Global Dialogue on AI Governance can play three specific roles that no other UN platform currently fills. First, a clearinghouse for open-source technical safeguards. The Dialogue should not attempt to negotiate binding treaties. That is the role of other UN bodies. Instead, it should curate and publish open-source tools – zero-knowledge verification protocols, API kill switches, cryptographic watermarking – that any State, humanitarian agency, or private sector actor can adopt immediately. These tools exist. They are not being systematically shared. Second, a capacity-building multiplier. The Dialogue should endorse and distribute zero-cost training resources for the 89 countries without AI expertise. I have drafted a field manual, "How to Train 100 Diplomats on AI for $0," using open-source models. The UN can translate, adapt, and distribute this manual at no cost. The Dialogue's value is not creating new content – it is amplifying what already works. Third, a convener for coalitions of the willing. Not every State is ready for binding AI governance. The Dialogue should create space for coalitions to form around specific issues – civilian digital infrastructure protection, biometric surveillance, AI explainability – without requiring consensus from all 193 member states. The Ottawa Treaty on landmines emerged outside the UN consensus process. The Dialogue can serve as the incubation space for similar coalitions. The Dialogue should measure its success not by declarations signed, but by tools adopted, diplomats trained, and coalitions formed.

The AI Dialogue should build upon four existing mechanisms. First, the UN Secretary-General's High-Level Advisory Body on AI. The Advisory Body identified the capacity gap and called for innovative financing. The Dialogue can operationalize this by distributing zero-cost training manuals and connecting donors (e.g., PGF AGI) with implementing partners (e.g., the Academy for AI Diplomacy and International Affairs). Second, the CCW Group of Governmental Experts on Lethal Autonomous Weapons Systems (LAWS). The LAWS process addresses military AI. The Dialogue addresses civilian AI. But there is overlap – civilian AI systems are repurposed for targeting. The Dialogue should explicitly refer this gap to the LAWS process, creating a formal handshake between civilian and military AI governance. Third, the International Committee of the Red Cross (ICRC) "digital emblem" initiative. The ICRC has called for innovative solutions to protect humanitarian data. The Dialogue can promote the ICRC's work and connect it with open-source technical safeguards from civil society. Fourth, the Mellow Foundation in Geneva. Mellow has launched an AI apprenticeship program for diplomats and international organization staff. The first cohort has graduated. The Dialogue can scale this model by encouraging other foundations to replicate it regionally. The added value of the Dialogue is not new initiatives. It is connectivity and enhanced cooperation. The Dialogue sits at the intersection of these efforts. Its role is to ensure they are not duplicating each other, and to create pathways for civil society innovations (like zero-cost training manuals) to reach UN agencies and member states.

Different stakeholders should contribute in different ways, based on their comparative advantage. Governments should articulate national priorities and identify red lines. The Dialogue should allocate dedicated time for regional groupings (Africa, Asia-Pacific, Latin America, Eastern Europe, Western Europe) to coordinate positions before plenary sessions. Civil society (including the Academy for AI Diplomacy and International Affairs) should submit concrete, implementation-ready tools – model frameworks, open-source technical annexes, zero-cost training curricula. The Dialogue should create a public repository for these submissions, organized by thematic area. Technical community (engineers, cryptographers, AI researchers) should provide reality checks on what is technically feasible versus politically desirable. The Dialogue should include dedicated "technical deep-dive" sessions alongside diplomatic plenaries. Private sector should demonstrate existing safeguards (e.g., API kill switches, watermarking) through live demonstrations, not just policy papers. Recommendations for format: Hybrid by default. Remote participation must be seamless, not an afterthought. Outcome-oriented working groups. Instead of open-ended discussions, each working group should produce a specific deliverable: a model law, a technical specification, a training curriculum. "Unconference" sessions. Half-day, participant-driven sessions where stakeholders propose topics on the spot and breakout groups form organically. Public repository of submissions. All written inputs should be published openly, with version control and commenting. The Dialogue should measure success by outputs, not attendance.

Five communities are systematically underrepresented. First, diplomats from the 89 countries without AI capacity. The UN Secretary-General's High-Level Advisory Body on AI identified this gap. These diplomats attend negotiations without technical briefings, relying on talking points from capital that may be outdated. Inclusion means providing pre-reads in multiple languages, zero-cost training (e.g., my manual "How to Train 100 Diplomats on AI for $0"), and dedicated mentorship from technically proficient missions. Second, humanitarian practitioners. ICRC, UNHCR, and WFP staff manage refugee data, food vouchers, and family reunification systems. They know how AI governance gaps affect vulnerable populations. Yet these voices are rare in AI governance forums. Inclusion means dedicated sessions at the Dialogue where humanitarians brief diplomats on operational realities. Third, small island developing States (SIDS) and least developed countries (LDCs). Their primary AI governance concern is not innovation – it is protection from AI-enabled cyber risks, disinformation, and data exploitation. Inclusion means regional pre-Dialogue consultations and travel support for delegates who cannot otherwise attend Geneva sessions. Fourth, independent treaty drafters and legal entrepreneurs. The UN system privileges institutional voices. But many innovative model frameworks (e.g., the Digital Hague Convention) come from outside the system. Inclusion means a formal "open call" for model frameworks, with a pathway for civil society submissions to be discussed in working groups. Fifth, open-source developers. The engineers who build zero-knowledge proofs, API kill switches, and watermarking tools rarely attend UN forums. Inclusion means virtual technical deep-dives and recognition that code is a form of governance.

Four innovative formats would foster meaningful engagement. First, a "Code as Governance" demo lab. One full day where developers demonstrate open-source technical safeguards – zero-knowledge verification, API kill switches, cryptographic watermarking – to diplomats and humanitarians. Live coding. No PowerPoint. Delegates leave with URLs to working code, not just declarations. Second, scenario-based negotiation simulations. Instead of abstract debates, run 90-minute simulations modeled on real incidents: a severed undersea cable, a breached humanitarian database, an AI-generated kill list. Delegates play assigned roles (State, ICRC, tech company). The simulation reveals governance gaps faster than any policy paper. Third, a "coalition of the willing" incubator. Dedicated space for groups of States to draft side agreements without requiring consensus. The Ottawa Treaty on landmines started this way. The Dialogue provides the venue, the facilitators, and the political cover. Fourth, a public repository with version control. All written submissions, working group outputs, and model frameworks should be published on a platform with commenting, forking, and version tracking – like GitHub for AI governance. This allows civil society submissions (like the Digital Hague Convention) to be cited, adapted, and improved by States without formal adoption. Fifth, "reverse panel" sessions. Instead of experts speaking to an audience, the audience sets the agenda. Attendees submit questions via app. Experts respond in real time. The moderator's job is to interrupt, redirect, and ensure panelists actually answer. The worst format is the traditional panel: four experts, 60 minutes, fifteen minutes for Q&A. The Dialogue should ban this format entirely.

I offer three examples, each at a different level of governance. Example 1: A model treaty - The Digital Hague Convention (2026) The Digital Hague Convention is a complete, binding treaty that prohibits using civilian data (mobile signals, biometrics, payment histories) for AI targeting in armed conflict and protects digital infrastructure (data centers, undersea cables, energy grids) as neutral assets. It operates at the level of international law. Its innovations include: A definition of AI "rubber-stamping" (≥80% human approval of AI recommendations with <30 seconds per review = autonomous weapon system) Corporate criminal liability for technology companies that knowingly enable prohibited targeting Open-source technical annexes (zero-knowledge verification, API kill switch, cryptographic watermarking) The treaty is complete. 27 articles, 6 annexes. Netherlands depositary. Signature opens Q1 2027. Full text available on request. Example 2: A zero-cost capacity-building tool - "How to Train 100 Diplomats on AI for $0" This field manual provides an 8-week, self-paced curriculum using open-source AI models (DeepSeek, Mistral, Google). No infrastructure. No budget. No IT staff. A diplomat anywhere can produce an intelligence briefing in 7 days. The manual is Creative Commons (CC BY-SA). The UN can translate, adapt, and distribute it at no cost. Example 3: A neutral convening platform - The Academy for AI Diplomacy and International Affairs (in formation) The Academy is a proposed Singapore-Geneva dual structure designed to drive ratification of the Digital Hague Convention, train diplomats using the zero-cost manual, and provide technical implementation support for open-source annexes. These three examples operate at different levels - treaty, training, implementation - but share a common philosophy: open-source, coalition-driven, and ready for adoption without waiting for universal consensus.