New York University - Peace Research and Education Program

The first Global Dialogue would succeed if it establishes concrete institutional mechanisms, not only declarative commitments, for ensuring that AI governance is shaped by those most affected by AI systems.Three outcomes would signal meaningful progress. First, a commitment to creating a formal submission pathway for community governed institutions to contribute governance models and evidence to future Dialogue sessions. Currently, the AI governance landscape is rich in consultation processes but structurally thin on mechanisms for transmitting governance innovations developed at the community level, particularly in the Global South, into multilateral policy formation. The Dialogue should address this by enabling submissions from community governed data institutions, indigenous governance bodies, and civil society organizations with equal procedural standing to government and private sector inputs.Second, the Dialogue should produce a clear mapping of how its work complements and connects to existing AI governance processes (the OECD AI Principles, regional frameworks such as the AU and ASEAN approaches, the EU AI Act) rather than duplicating them. The Dialogue's added value lies precisely in its universality and its multistakeholder mandate; using this to build interoperability between governance approaches, including community level and indigenous governance frameworks, would be a distinctive and urgently needed contribution.Third, the Dialogue should commit to establishing a repository or clearinghouse for participatory AI governance models, hosted by the joint Secretariat. Communities in Malawi, the Philippines, Kenya, and many other countries are already building governance architectures that address accountability, consent, and benefit sharing in context specific ways. A curated, accessible repository would provide Member States with practical examples to draw upon while recognizing the intellectual contribution of communities who developed them.Success, in short, means moving from consultation to co-design: building the institutional infrastructure for a genuinely inclusive AI governance process, not only convening one.

• AI capacity-building
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

These four priorities reflect the areas where the gap between international ambition and operational reality is widest, and where participatory governance approaches offer the most immediate practical contribution.AI capacity building is foundational, but it must be understood expansively: not only as technical training or compute access, but as the ability of communities to govern AI systems on their own terms. In our work across Malawi, the Philippines, and Kenya, capacity building means equipping communities with governance frameworks, not only technical skills. The Malawi Voice Data Commons, for instance, required building institutional capacity for traditional authorities to exercise oversight over how voice data is collected, stored, and used in AI applications.Interoperability of governance approaches is urgent precisely because the most innovative governance models are emerging at the community level and have no pathway into international frameworks. Indigenous governance systems such as the Bodong in the Philippines offer sophisticated, time tested approaches to collective decision making, consent, and reciprocity that are directly relevant to AI governance, yet they remain invisible to multilateral processes.The social, economic, ethical, cultural, and linguistic implications of AI cannot be assessed from a distance. Our research in Kenya's informal settlements documents how AI driven platforms restructure local economies and concentrate informational power. These implications are experienced differently across contexts, and governance responses must reflect that specificity.Finally, open source software, open data, and open AI models are essential infrastructure for digital sovereignty, but openness without governance reproduces existing power asymmetries. A commons based approach, where openness is paired with community authority over terms of use and benefit sharing, offers a more sustainable model. Our partnership with Mozilla's data collective platform in the Philippines illustrates how open infrastructure can be governed through indigenous institutional frameworks.

Two cross-cutting issues deserve explicit attention in the Dialogue's thematic architecture. First, data sovereignty as a governance question. The listed themes address AI systems and their implications, but none directly names the question of who governs the data infrastructures on which AI systems depend. Data is the backbone of AI governance: decisions about data collection, storage, access, and use determine who benefits from AI, who bears its risks, and who has standing to shape its development. Indigenous communities, rural populations, and speakers of underrepresented languages face particular risks of data extraction without meaningful consent or reciprocal benefit. The CARE Principles for Indigenous Data Governance (Collective Benefit, Authority to Control, Responsibility, Ethics) provide a tested, rights based framework for addressing this gap, and the Dialogue should engage with them explicitly.Second, the concentration of AI related power in specific geographies and institutions. The current governance discourse tends to frame the challenge as a "digital divide" to be bridged through capacity building and technology transfer. While important, this framing can obscure a more structural dynamic: AI systems actively concentrate economic, informational, and decision making power in ways that reshape local governance, labour markets, and knowledge systems. Our research in Nairobi's informal settlements shows that AI driven platforms do not simply bypass underserved communities; they reorganize local power relations in ways that require governance responses, not only connectivity solutions. The Dialogue should address AI related power concentration as a governance challenge in its own right, distinct from and complementary to the capacity building agenda. Both issues cut across the listed themes and would benefit from dedicated analytical attention in the July programme, whether through a cross-cutting framing document prepared by the Scientific Panel or through a dedicated thematic session.

Our research programme operates across three regions of the Global South (southern Africa, Southeast Asia, and East Africa), and in each, the governance gaps we have identified manifest in strikingly similar ways despite very different political and institutional contexts. The most significant challenge is what we call the extraction without governance problem. AI systems are being trained on data drawn from communities that have no institutional mechanism to shape how that data is used, who profits from it, or what obligations flow back. In Malawi, Chichewa is spoken by over 15 million people, yet speakers have had no role in governing the speech data that feeds commercial AI language technologies. In the Philippines, indigenous knowledge systems are increasingly legible to AI scraping tools, but existing intellectual property and data protection frameworks were not designed to recognize collective, intergenerational knowledge holders. In Kenya, AI driven platforms operating in informal settlements extract transactional and behavioral data that reshapes local economic structures, with no governance layer through which affected communities can exercise oversight. The opportunity, however, is equally significant. In all three contexts, communities have responded not by rejecting AI but by building governance architectures that assert local authority. Traditional authorities in Malawi have co-designed a three tier governance infrastructure for voice data. Kalinga communities in the Philippines are adapting the Bodong peace pact system to govern AI's interaction with indigenous knowledge. Community researchers in Nairobi are producing governance recommendations grounded in lived experience of AI's effects. These innovations demonstrate that participatory AI governance is not aspirational; it is already operational. The governance gap lies not in the absence of community capacity but in the absence of multilateral mechanisms to recognize, support, and learn from what communities are already building.

The AI Dialogue's most important contribution to international cooperation would be to address a structural problem that existing fora have not resolved: the absence of a mechanism for governance innovations developed at the community level to inform multilateral AI policy.Current international cooperation on AI governance operates primarily through intergovernmental channels (the OECD, G7, G20) and multistakeholder platforms that, while valuable, tend to privilege inputs from governments, large technology companies, and established research institutions. The result is a governance conversation that is global in aspiration but narrow in sourcing. Communities building sophisticated AI governance architectures in the Global South, including indigenous governance bodies, community governed data institutions, and locally embedded civil society organizations, have no structured pathway to contribute their models and evidence to international processes.The AI Dialogue, by virtue of its universal membership and its explicit multistakeholder mandate under Resolution 79/325, is uniquely positioned to fill this gap. Three functions would be particularly valuable. First, the Dialogue could serve as a translation layer between community level governance practice and intergovernmental policy discussion, ensuring that locally developed models are documented, assessed, and made available to Member States. Second, it could advance interoperability not only between national regulatory frameworks but between different scales and traditions of governance, including indigenous and customary systems that predate and complement state based regulation. Third, it could provide a platform for South to South exchange on AI governance, enabling communities and researchers working on similar challenges in different regions to share approaches without routing through Northern intermediaries.International cooperation on AI will remain incomplete as long as it is structured exclusively around what governments and corporations bring to the table. The Dialogue should be the space where the rest of the world's governance capacity becomes visible to multilateral processes.

The AI Dialogue should connect with and build upon several categories of existing work rather than duplicating them. At the multilateral level, the OECD AI Policy Observatory, the Global Partnership on AI, and the Council of Europe Framework Convention on AI provide important normative and analytical foundations. The Dialogue should position itself as complementary to these efforts, focusing on universality and multistakeholder depth where existing fora are limited by membership or mandate. The work of UNESCO, particularly the Recommendation on the Ethics of AI adopted in 2021 and the ongoing efforts of expert groups such as Women4EthicalAI, offers a normative framework the Dialogue should reference rather than reinvent. At the regional level, the African Union's Continental AI Strategy, ASEAN's Guide on AI Governance and Ethics, and the EU AI Act represent important but varied approaches. The Dialogue's added value lies in fostering interoperability across these frameworks, identifying common principles while respecting contextual specificity.Perhaps most critically, the Dialogue should build upon community level governance initiatives that rarely appear in multilateral inventories but represent some of the most advanced operational work on participatory AI governance. These include commons based data governance projects (such as the Malawi Voice Data Commons and similar initiatives in other regions), indigenous data sovereignty networks working with the CARE Principles, and locally embedded research partnerships that produce governance recommendations grounded in lived experience. The Mozilla Foundation's data collective infrastructure, which hosts community governed data initiatives, and the Global Indigenous Data Alliance offer institutional entry points for this engagement.The Dialogue's added value is not to replicate what these initiatives do but to create the connective tissue between them: a space where community governance models, regional regulatory frameworks, and multilateral norms can interact, learn from each other, and collectively shape an AI governance architecture that is genuinely global in scope and in sourcing.

Different stakeholders bring different forms of knowledge to AI governance, and the Dialogue's structure should reflect this rather than defaulting to a single format that privileges polished institutional statements. Member States bring regulatory authority and policy implementation experience. The Dialogue should create space for them to share not only frameworks but candid assessments of what is and is not working in national AI governance, particularly around enforcement, institutional coordination, and resource constraints. The private sector brings technical knowledge and implementation capacity, but also bears particular accountability for AI's effects. The Dialogue should move beyond general commitments to responsible AI and create structured formats where companies engage with specific governance proposals from affected communities. This means designing sessions where private sector representatives respond to community evidence rather than presenting alongside it. Civil society and academia bring independent analysis, participatory methodologies, and long term engagement with affected communities. Our research across Malawi, the Philippines, and Kenya demonstrates that academic institutions embedded in local contexts can serve as institutional bridges between community governance innovation and international policy processes. The Dialogue should formally recognize and resource this bridging function. Most critically, indigenous peoples, local communities, and practitioners from the Global South bring governance knowledge that is operational, tested, and grounded in collective decision making traditions. Their contribution should not be reduced to "perspectives" offered during consultations; it should include the submission of governance models and institutional designs that the Dialogue formally receives and makes available to Member States. On structure, I recommend the Dialogue adopt a layered format: plenary sessions for high level framing, followed by thematic working sessions where stakeholders engage with specific governance proposals rather than delivering sequential statements. Evidence based exchanges produce better policy than speaker lists.

Three categories of voices remain structurally underrepresented in global AI governance discussions, not because they lack expertise but because existing institutional formats were not designed to include them. First, indigenous peoples and traditional governance authorities. Indigenous communities are among those most directly affected by AI driven data extraction, yet global AI governance discussions rarely engage with indigenous governance systems as sources of institutional design. The Bodong system in the Philippines, customary authority structures in Malawi, and community governance practices across East Africa offer sophisticated models for collective decision making, consent, and reciprocity that are directly relevant to AI governance. Inclusion requires more than invitations to speak; it requires recognizing indigenous governance bodies as institutional contributors with standing to submit governance models and frameworks. Second, communities in informal and underserved settings in the Global South. Our research in Nairobi's informal settlements shows that AI systems are actively reshaping local economies and power relations, yet residents of these communities are almost entirely absent from governance discussions at the international level. Inclusion requires funded participation pathways and formats that allow community researchers to present evidence on their own terms, not filtered through institutional intermediaries. Third, speakers of underrepresented languages. AI governance discussions are conducted overwhelmingly in English, and AI systems are being built primarily for high resource languages. Communities whose languages are being incorporated into AI training datasets (often without consent or governance mechanisms) should have a recognized role in discussions about linguistic and cultural implications of AI. The Dialogue should provide interpretation in languages beyond the six UN official languages and create written submission pathways that accept inputs in local languages. Meaningful inclusion is structural, not ceremonial. It requires funded travel, accessible formats, and governance standing, not only speaking slots.

The most significant barrier to meaningful engagement in multilateral AI governance processes is not a lack of goodwill but a reliance on formats that structurally favour institutional actors with prepared statements. Three innovations could shift this dynamic. First, evidence exchange sessions modelled on participatory research practice. Rather than sequential three minute statements, thematic sessions should be structured around specific governance challenges, with stakeholders responding to concrete case studies or governance proposals. In our workshops across Malawi, the Philippines, and Kenya, we have found that stakeholders engage far more substantively when responding to a specific scenario (for example: how should consent be governed when voice data from a rural community is used to train a commercial speech recognition model?) than when delivering pre-prepared remarks on broad themes. The Dialogue could adapt this format by commissioning short governance challenge briefs and structuring discussions around them. Second, community evidence submissions with formal Dialogue standing. The Dialogue should establish a mechanism for communities and civil society organizations to submit governance models, case studies, and policy recommendations through a curated portal. These submissions should be synthesized by the Secretariat and presented as formal inputs alongside Member State and private sector contributions, ensuring they inform the Co-Chairs' summary and future programme design. Third, regional and thematic pre-sessions conducted in partnership with local institutions. Rather than concentrating all engagement in Geneva, the Dialogue should commission preparatory workshops in the Global South, hosted by universities, civil society organizations, and community institutions that can convene locally relevant stakeholders. The outputs of these sessions should feed directly into the July programme. This approach would also address the structural exclusion created by travel costs, visa barriers, and time zone constraints. Format shapes substance. If the Dialogue innovates on how people participate, it will produce better governance outcomes than if it only innovates on what they discuss.

Three categories of examples illustrate how effective AI governance can be built from the ground up, not only from the top down. At the community governance level, the Malawi Voice Data Commons provides a concrete model for how affected communities can exercise collective authority over AI related resources. The project established a three tier governance infrastructure: community assemblies led by traditional authorities make decisions about data collection and use; university partners (MUBAS, MUST, Mzuzu University) provide technical stewardship; and a coordinating body ensures accountability between the two. This architecture ensures that AI speech technologies built from Chichewa voice data operate under terms set by speakers themselves. In the Philippines, the Indigenous Knowledge Data Collaborative at Kalinga State University adapts the Bodong, an indigenous peace pact governance system, to regulate how AI systems may interact with collective indigenous knowledge. The Bodong framework addresses consent, reciprocity, and benefit sharing through governance mechanisms that predate and complement state regulation. Both projects are hosted on Mozilla Foundation's data collective platform, demonstrating that open technical infrastructure and community governance authority are not only compatible but mutually reinforcing. At the policy framework level, the CARE Principles for Indigenous Data Governance (Collective Benefit, Authority to Control, Responsibility, Ethics) offer a rights based framework that translates indigenous governance values into operational guidelines applicable across diverse legal contexts. These principles are increasingly referenced in national data strategies and provide a model for how the Dialogue could integrate community governance norms into international AI policy discussions. At the research and evidence level, our community led research in Nairobi's informal settlements demonstrates a participatory methodology where residents themselves document AI's effects on local power structures and produce governance recommendations. This approach generates policy evidence that is both rigorous and grounded in lived experience, a model the Dialogue could scale through partnerships with locally embedded academic institutions.