Geoblood

Success would mean the Dialogue produces concrete, actionable commitments rather than another layer of principles that developing nations cannot implement. From the perspective of a nonprofit deploying AI in public health emergencies across underserved communities, three outcomes matter most.First, a shared framework for AI in humanitarian and health applications that distinguishes between high-risk commercial AI and mission-driven AI deployed for public good. Current governance discussions treat all AI uniformly, creating compliance burdens that disproportionately disadvantage small nonprofits and civil society organizations serving the most vulnerable populations.Second, meaningful capacity-building commitments with timelines. The countries facing the most acute AI governance gaps are the same countries where AI could have the greatest humanitarian impact. Governance frameworks must be designed with implementation capacity in mind, not just aspirational standards that only well-resourced actors can meet.Third, a recognition that AI governance is inseparable from AI access. Populations in South Asia, sub- Saharan Africa, and the Middle East are already subject to AI-driven decisions without meaningful participation in shaping the rules. The Dialogue succeeds when those communities gain both governance voice and practical access to beneficial AI tools.Success is not a communiqué. It is a roadmap that a health nonprofit in Missouri or a blood bank in Lagos can actually use.

• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Open-source software, open data and open AI models

GeoBlood is a US-registered 501(c)(3) nonprofit building AI-powered blood donor-recipient matching infrastructure for communities where blood shortages cause preventable deaths. Our work operates at the intersection of public health, emergency response, and AI - in geographies ranging from Missouri to South Asia and sub-Saharan Africa. These four priorities reflect the governance gaps we encounter directly. AI capacity-building is our most urgent priority. The communities GeoBlood serves - in low- and middle-income countries where blood shortages are most severe - lack the technical infrastructure, regulatory frameworks, and trained practitioners to safely deploy or oversee AI systems. Governance without capacity is exclusion by another name. The social, economic, ethical, cultural, and technical implications of AI are inseparable from our mission. Algorithmic donor matching decisions affect who receives life-saving blood first. Embedding cultural sensitivity, linguistic accessibility, and equity into AI design is not optional in our context - it is the difference between a tool that saves lives and one that replicates existing health inequities. Human rights protection is foundational. Health data is among the most sensitive personal information that exists. GeoBlood operates under HIPAA-aware architecture, but comparable protections do not exist in many of our target geographies. International human rights standards applied to health AI would create a baseline that protects our users regardless of where they live. Open-source models and open data directly enable our mission. GeoBlood is building an open-source AI foundation model for blood supply gap prediction that any health system globally can adopt and adapt. International norms encouraging open AI development in the public health domain would accelerate the adoption of tools like ours in the communities that need them most.

Two critical issues are absent from the listed themes. The first is AI in emergency and humanitarian response. The Dialogue's thematic framework addresses AI implications broadly but does not specifically address the governance of AI systems deployed in life-or-death emergency contexts. When an AI platform is making real-time decisions about which blood donors to alert during a medical emergency, the governance requirements are fundamentally different from those governing a recommendation algorithm or a content moderation system. Emergency AI requires specific standards for speed, reliability, bias auditing in high-stakes conditions, and accountability when the system fails. This gap leaves organizations like GeoBlood - and more broadly, humanitarian AI deployments in disaster response, disease outbreak management, and emergency medical coordination - without governance guidance tailored to their operational realities. The second is the governance of AI developed by civil society and nonprofits for public good. Virtually all current AI governance discourse is framed around commercial actors and state actors. Civil society organizations deploying AI for humanitarian purposes occupy a distinct governance space: they are mission-driven rather than profit-driven, they serve populations that commercial actors ignore, and they often operate across multiple jurisdictions with inconsistent regulatory requirements. A specific governance track for nonprofit and humanitarian AI - one that preserves the open, community-first principles that make this work valuable while providing appropriate accountability frameworks - would fill a significant gap in the current landscape. Both issues are urgent. As AI capabilities expand into health emergency response and humanitarian logistics, the absence of tailored governance frameworks creates real risks for the world's most vulnerable populations.

GeoBlood operates at the intersection of public health AI and humanitarian emergency response across North America, South Asia, and the Middle East — precisely the regions where governance gaps cause the most tangible harm. In the United States, GeoBlood's home jurisdiction, health AI governance remains fragmented across federal agencies with no unified framework governing AI in emergency medical coordination. This creates uncertainty for mission-driven nonprofits attempting to deploy AI responsibly. We have invested significantly in HIPAA-aware data architecture and bias auditing despite having no regulatory requirement to do so, because the absence of clear standards forces each organization to define its own accountability floor independently. In South Asia and sub-Saharan Africa, where blood shortages are most acute and GeoBlood's impact potential is highest, the governance gap is more severe. India has a 41 million unit annual blood shortfall. Yet health AI governance frameworks in these regions are nascent at best, leaving communities simultaneously most exposed to AI-driven health decisions and least equipped to shape or challenge them. Local health authorities lack the technical capacity to evaluate AI systems, creating conditions where harmful or poorly designed tools can proliferate unchecked alongside beneficial ones. The most significant opportunity is that these same governance gaps represent a window for civil society to help define the standards before commercial interests do. GeoBlood's open-source AI foundation model for blood shortage prediction is designed to be the kind of public-good infrastructure that governance frameworks should be incentivizing. If the Dialogue can establish norms that favor open, community-verified, mission-driven AI in public health settings, it would directly accelerate the adoption of tools that save lives in the communities most underserved by existing health infrastructure. Governance is not only a constraint. Designed well, it is the mechanism that ensures AI reaches the people who need it most.

he AI Dialogue occupies a unique position in the international governance landscape: it is the only multilateral forum explicitly designed to bring governments, civil society, academia, and technical experts into the same room on equal footing. That structural inclusivity is its greatest asset and its most important contribution to international cooperation. The most valuable role the Dialogue can play is not producing binding rules — that process will take years and risks being captured by the actors with the most lobbying capacity. Instead, the Dialogue can build the shared understanding that makes eventual binding frameworks both possible and legitimate. Countries cannot cooperate on AI governance if their technical experts, policymakers, and civil society organizations are operating from fundamentally different assumptions about what AI does, how it fails, and who it harms. For organizations like GeoBlood, operating across multiple jurisdictions with inconsistent regulatory environments, the practical need is for interoperable governance frameworks that do not require a different compliance architecture for every country we enter. The Dialogue can advance this by establishing minimum shared standards for AI transparency, bias auditing, and accountability in high-stakes domains like health and emergency response — standards that national frameworks can adopt and adapt rather than invent independently. The Dialogue also has a unique convening role for the Global South. Many low- and middle-income countries have sophisticated perspectives on AI's risks and opportunities but lack the institutional presence in existing governance forums to make those perspectives heard. A genuinely inclusive Dialogue would surface governance insights from health systems in Lagos, blood banks in Karachi, and emergency responders in Nairobi that no amount of expert consultation in Geneva can replicate. Cooperation at this level does not happen automatically. The Dialogue must actively create the conditions for it.

Several existing frameworks provide foundations the Dialogue should connect with rather than duplicate. The UNESCO Recommendation on the Ethics of AI, adopted by 193 member states, represents the broadest existing consensus on AI principles and provides a human rights baseline the Dialogue should explicitly build upon rather than renegotiate. The OECD AI Principles, while primarily representing high-income countries, offer a practical implementation framework that has already influenced national legislation in dozens of jurisdictions. The ITU's AI for Good platform, which hosts the Summit alongside which the Dialogue convenes, has an existing network of humanitarian AI projects and civil society organizations that the Dialogue should systematically engage. These are the practitioners who understand where governance frameworks succeed and fail in real-world deployment conditions. The Global Digital Compact's commitments on digital inclusion and open AI infrastructure provide a mandate the Dialogue should operationalize, particularly around ensuring that open-source AI tools for public health and humanitarian response receive explicit governance support rather than being treated identically to proprietary commercial systems. For health-specific AI, the WHO's guidance on ethics and governance of AI for health represents domain expertise the Dialogue should formally connect with. GeoBlood's experience suggests that health AI governance requires sector-specific standards that general frameworks cannot adequately address. The added value the Dialogue uniquely brings is legitimacy and universality. The frameworks above are either non-binding, regionally limited, or sector-specific. The Dialogue, operating under UN auspices with participation from all member states and civil society, can convert the best elements of existing frameworks into shared reference points that carry the weight of genuine multilateral consensus.

The AI Dialogue's structure should reflect the reality that AI governance expertise does not reside exclusively in governments or large institutions. Different stakeholders bring irreplaceable knowledge that formal diplomatic formats systematically exclude. Governments bring regulatory authority and implementation capacity but often lack technical depth and community-level insight. Their role should be committing to implementation timelines, not defining technical standards unilaterally. Civil society organizations, particularly those deploying AI in humanitarian and public health contexts, bring ground-truth evidence of how governance gaps cause real harm. GeoBlood's experience navigating inconsistent health data regulations across four continents represents exactly the kind of practitioner knowledge that should inform governance standards. Civil society should have structured input mechanisms with guaranteed response, not merely observer status. Technical experts and academic institutions should be tasked with translating governance principles into implementable standards, with explicit mandates to make those standards accessible to resource-constrained organizations in low- and middle-income countries. The private sector should participate with transparency requirements. Any company contributing to governance discussions should disclose the commercial interests their proposals would advance or protect. Structurally, the Dialogue should avoid the UN's tendency toward plenary-dominated formats where power differentials are most pronounced. Smaller working groups organized by sector and impact domain, with rotating facilitation, would produce more substantive outputs. Between annual convenings, a maintained digital platform for asynchronous written contribution would allow participation from organizations that cannot afford Geneva travel, ensuring the Dialogue's inputs reflect genuine global diversity rather than the subset of actors with international conference budgets. Most importantly, the Dialogue should publish its inputs transparently and explain how they influenced outputs. Without that accountability loop, contributing feels performative.

The most consequential gap in current AI governance discussions is the near-total absence of communities that are already subject to AI-driven decisions without any meaningful representation in the forums shaping the rules. Communities in South Asia, sub-Saharan Africa, and the Middle East face the world's most acute AI governance challenges — algorithmic systems making health, credit, and social welfare decisions — while being systematically underrepresented in Geneva, New York, and Brussels. Language barriers, travel costs, and institutional access requirements create a governance process that is structurally biased toward the actors with the least urgent need for protection. Frontline health workers in low-income settings represent another invisible constituency. GeoBlood's experience working with community health organizations in underserved geographies consistently surfaces governance insights that academic literature misses entirely — practical concerns about connectivity reliability, cultural trust dynamics, and the real-world failure modes of AI matching systems that no amount of theoretical ethics analysis anticipates. Indigenous communities deserve explicit inclusion. AI systems trained on non-representative data perpetuate historical exclusions. Indigenous data sovereignty frameworks, developed outside formal governance forums, offer governance models the Dialogue should learn from rather than overlook. Inclusion mechanisms must go beyond translation. Dedicated funding for participation from underrepresented regions, structured pre-consultation processes that surface community perspectives before formal sessions, and governance roles with actual decision influence rather than ceremonial presence are minimum requirements. The Dialogue should measure its inclusivity not by the number of countries represented but by whether the communities most harmed by ungoverned AI had their specific concerns reflected in the outputs. That standard has never been met by any international AI governance forum. The inaugural Global Dialogue has the opportunity to be the first.

The most persistent failure of international governance forums is that their formats optimize for diplomatic procedure rather than genuine knowledge transfer. The AI Dialogue should experiment with formats explicitly designed to surface practitioner knowledge and generate actionable outputs. Reverse expert panels would invert the typical structure. Rather than technical experts presenting to policymakers, frontline practitioners from health systems, humanitarian organizations, and community groups in low-income settings present the governance failures they have experienced directly, with technical experts responding to those specific cases. This format grounds abstract governance discussions in operational reality. Red team sessions would assign participant groups to stress-test proposed governance frameworks against specific high-stakes scenarios — an AI blood matching system failing during a disaster, a facial recognition system misidentifying aid recipients, an algorithmic credit system excluding smallholder farmers. Stress-testing against real scenarios produces more durable governance frameworks than principles-only discussions. Asynchronous digital tracks running parallel to the in-person Dialogue would allow organizations that cannot attend physically to contribute substantively. GeoBlood's team spans the UAE, Pakistan, and the United States. Meaningful participation in a Geneva event requires a digital engagement pathway that is not a passive comment form but an active structured input mechanism with visible influence on session agendas. Outcome accountability sessions at the close of each Dialogue day should ask explicitly: what specific commitment is each stakeholder group making before the next Dialogue? Publishing those commitments publicly and reviewing progress at the following year's session creates an accountability loop that most governance forums deliberately avoid. The Dialogue's legitimacy will ultimately be measured by whether participating organizations change their behavior as a result. Formats that generate commitments rather than communiqués are the only ones worth designing.

Several concrete examples demonstrate what effective AI governance looks like when it moves beyond principles into practice. GeoBlood's own approach offers a replicable model for mission-driven AI governance in resource-constrained settings. Before deploying any AI matching component, we built a HIPAA-aware data architecture that treats privacy protection as infrastructure rather than compliance. We conduct bias audits at each model iteration specifically testing whether the system performs equitably across blood types, demographic groups, and geographic contexts. We have committed to releasing our foundation model under open-source license so that any health system globally can audit, validate, and adapt it. This is not regulatory compliance. It is governance by design, built into the development process from the first line of code. The European Union's AI Act provides the most comprehensive example of risk-tiered governance, distinguishing between high-risk AI applications in health and safety contexts and lower-risk applications with proportionate requirements. While its implementation burden is significant for small organizations, its underlying logic - that governance requirements should scale with potential harm - is the correct framework and should inform the Dialogue's approach to sector-specific standards. The WHO's Ethics and Governance of AI for Health guidance demonstrates how domain-specific frameworks can translate general principles into actionable standards for practitioners. Its emphasis on transparency, accountability, and the protection of vulnerable populations directly addresses the gaps that general AI governance frameworks leave open. Mozilla's Responsible AI Challenge and similar open civil society initiatives demonstrate that governance innovation does not require intergovernmental agreement. Community-driven accountability standards, developed by the organizations closest to deployment realities, often outpace formal regulatory frameworks in both speed and practical relevance. The shared lesson across these examples is that effective governance is specific, measurable, and built by the people closest to the consequences.