Asuntos del Sur

A successful first Global Dialogue on AI Governance would produce three concrete outcomes. First, agreement on a multi-stakeholder framework that formally recognises civil society, academia, and technical communities — not only states and industry — as legitimate actors in AI governance processes. Second, the adoption of human rights as the binding normative foundation for any future international instrument, ensuring that principles such as non-discrimination, transparency, and accountability are not treated as optional add-ons but as structural requirements. Third, the establishment of a clear and inclusive roadmap for subsequent negotiations, with dedicated mechanisms to ensure that the perspectives of countries in the Global South are systematically integrated — rather than consulted as an afterthought. Success should not be measured solely by the breadth of consensus reached, but by whether the process itself models the inclusive, rights-respecting governance it seeks to promote. A dialogue that sidelines civil society or reproduces existing power asymmetries in AI development would undermine its own legitimacy from the outset.

• AI capacity-building
• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Protection and promotion of human rights

These four thematic areas are interconnected and, taken together, address the fundamental governance deficit that currently characterises the global AI landscape. Safe, secure and trustworthy AI and the protection and promotion of human rights form the normative core: without binding commitments to accountability, transparency, and non-discrimination, any governance architecture risks becoming a legitimising façade for the interests of the most powerful actors in AI development. Rights frameworks provide the only existing international consensus capable of grounding enforceable standards. AI capacity-building is not a peripheral concern but a prerequisite for meaningful participation. Countries and civil society organisations in the Global South cannot engage substantively in governance processes if the technical, institutional, and financial gap between them and frontier AI actors continues to widen. Capacity-building must therefore be understood as a governance instrument, not merely a development objective. Interoperability of governance approaches matters because no single framework will emerge from this process. What is achievable is a set of shared principles and mutual recognition mechanisms that allow diverse regulatory approaches to coexist without creating arbitrage opportunities for actors seeking the weakest oversight. Across all four areas, the legitimacy of long-term governance processes depends on whether they are genuinely shaped by diverse actors - including civil society, academic institutions, and governments from the Global South - or whether participation remains nominal. Inclusive processes are not merely more equitable; they are more robust, better at anticipating harms, and more likely to produce durable agreements. This is the lens through which Asuntos del Sur engages with the Dialogue.

Three cross-cutting issues deserve explicit attention in the Dialogue's agenda. First, AI infrastructure and security remain a little absent from the thematic framework: governance discussions cannot be decoupled from questions of who owns and controls the physical and digital infrastructure underpinning AI systems-data centres, semiconductor supply chains, cloud platforms, and foundational models, and specific considerations of their environmental impact. Without addressing these dependencies, governance frameworks risk regulating outputs whilst leaving the structural conditions of AI development untouched. Second, regional governance strategies represent a gap that the current thematic architecture does not adequately capture. Several regions-including Latin America and the Caribbean-are developing their own normative approaches and multi-country cooperation mechanisms that are not merely implementation vehicles for global frameworks but sites of genuine governance innovation that should inform, and not simply receive, international standards. Third, and perhaps most critically, the process must incorporate structural counterweights to hegemonic governance models. The current global AI landscape is shaped by a small number of states and corporations whose interests do not automatically align with those of the majority of the world's population. A legitimate governance process must therefore include formal mechanisms-not merely consultative ones-that allow a genuine plurality of actors to contest and shape governance decisions, including civil society organisations and Global South governments in substantive, decision-making roles. Taken together, these three dimensions point to the same underlying challenge: that the governance of AI cannot be separated from broader questions of power, sovereignty, and democratic legitimacy at the global level.

The governance gaps identified above have concrete and asymmetric consequences for civil society organisations and countries in Latin America and the Caribbean. The most pressing challenge is the systematic exclusion of diverse voices from the spaces where AI governance is actually being shaped. Global debates are currently dominated by a narrow set of actors — a handful of technologically advanced states and a small number of large technology corporations — whose framing of AI risks, opportunities, and regulatory solutions does not reflect the realities or priorities of the majority of the world's population. Civil society organisations working on democratic innovation, human rights, and public interest technology in the Global South are structurally absent from these conversations, not because they lack expertise or legitimate perspectives, but because the financial and institutional resources required to participate meaningfully are unevenly distributed. A lack of substantive, forward-looking debate at the regional level compounds this exclusion. Discussions in Latin America and the Caribbean tend to be reactive — responding to frameworks developed elsewhere — rather than propositional. There is insufficient investment in building the regional research, policy, and civil society capacity needed to articulate autonomous governance positions and engage on equal footing in international processes. The opportunity, however, is real. The region has accumulated significant experience in democratic innovation, open government, and multi-stakeholder participation, all of which are directly relevant to AI governance. If the Global Dialogue creates genuine space for these contributions — through dedicated funding, inclusive participation mechanisms, and a willingness to engage with governance models that challenge existing power asymmetries — Latin America and the Caribbean can move from being recipients of global AI governance frameworks to active co-designers of them.

The AI Dialogue has a rare and time-sensitive opportunity to establish the institutional foundations for a genuinely multilateral governance architecture — one that moves beyond declarations and towards mechanisms with real coordinative and accountability functions. To fulfil this role, the Dialogue must do more than produce consensus documents; it must catalyse the construction of a true multi-stakeholder network with standing participation structures, shared monitoring frameworks, and enforceable commitments that go beyond voluntary adherence.This requires a deliberate shift in how international cooperation on AI is conceived. Currently, cooperation tends to occur bilaterally or within clubs of technologically advanced states, producing frameworks that are then presented to the rest of the world as fait accompli. The Dialogue can interrupt this dynamic by creating permanent, inclusive spaces where civil society, technical communities, and governments from the Global South are not consulted after decisions are made, but participate in making them. Effective governance also demands enforcement: cooperation without accountability mechanisms produces norms that powerful actors can selectively ignore. The Dialogue should therefore explore binding commitments, peer review processes, and dispute resolution mechanisms that give international AI governance real institutional weight. Without these, the risk is that the process legitimises existing asymmetries rather than correcting them.

The AI Dialogue should actively build upon existing regional and global initiatives rather than constructing parallel architectures from scratch. In Latin America and the Caribbean, the eLAC framework — the regional digital agenda coordinated through ECLAC — represents a consolidated institutional platform with legitimacy across governments, civil society, and technical communities, and provides a natural entry point for translating global governance commitments into regional policy action. At the project level, initiatives such as LatamGPT demonstrate that regional cooperation on AI development and infrastructure is not only desirable but feasible. The Dialogue should recognise and connect with these efforts, supporting the conditions — technical, financial, and normative — under which regional AI initiatives can scale and diversify. Anchoring governance discussions to concrete cooperative projects of this kind helps ensure that frameworks remain grounded in operational realities rather than abstract principles. The Dialogue should also draw on the knowledge and convening capacity of global civil society networks such as Southern Voice and People Powered, which have developed robust models for amplifying the perspectives of organisations and communities systematically underrepresented in international policy processes. These networks offer not only thematic expertise but tested methodologies for inclusive participation that the Dialogue's own process design could directly adopt. The added value the Dialogue can bring to this ecosystem is integration: creating the connective tissue between regional initiatives, civil society networks, and intergovernmental processes that currently operate in silos, and providing the institutional legitimacy needed to translate their collective work into durable governance commitments.

Meaningful stakeholder contribution requires structural design choices, not merely open invitations. The Dialogue should move beyond panel-based formats — where participation is largely observational — towards co-creation spaces in which civil society, technical communities, governments, and affected populations jointly develop proposals, identify gaps, and stress-test governance options. This implies dedicated working sessions with substantive decision-making functions, not only opportunities to speak on the record.Participation must be genuinely hybrid, with virtual modalities designed to the same standard as in-person engagement — not as a fallback option but as a primary channel for organisations and communities that cannot bear the financial and logistical costs of international travel. Accessibility in language, format, and timing is a governance choice, and the Dialogue's credibility will partly depend on whether these choices reflect its stated inclusivity commitments.The Dialogue should also incorporate structured one-to-one engagement between civil society organisations and government delegations. Multilateral formats tend to flatten the specificity of bilateral exchange; dedicated spaces for direct dialogue between non-state actors and national representatives would allow for more substantive engagement on country-specific contexts, priorities, and constraints, and would help build the sustained relationships that durable governance cooperation requires.

The most significant representational gap in global AI governance is not merely geographic, though that dimension is real and consequential. It is structural: the voices of civil society organisations, grassroots movements, and communities that bear the most acute risks of AI-related harm are systematically excluded from the spaces where governance frameworks are designed. This includes organisations working directly with marginalised and vulnerable communities — indigenous peoples, racialised minorities, persons with disabilities, informal workers affected by algorithmic management, and communities subject to automated public administration systems — whose lived experience of AI's harms constitutes irreplaceable governance knowledge. It also includes grassroots and community-based organisations that lack the institutional resources to engage in international processes but whose proximity to affected populations makes their perspectives essential. Inclusion requires more than access: it requires investment. This means dedicated funding for participation, capacity-building support for organisations navigating international policy spaces for the first time, translation and interpretation beyond the major UN languages, and — critically — governance mechanisms that give these voices genuine influence over outcomes, not merely the opportunity to be heard. The Dialogue should establish explicit representation targets and accountability mechanisms to ensure that underrepresented communities move from the margins to the centre of the process.

Standard panel and plenary formats privilege established actors, reward institutional fluency, and tend to reproduce existing hierarchies rather than generate new insights. The Dialogue should instead draw on a broader repertoire of participatory and deliberative methodologies that have been developed — and rigorously tested — in democratic innovation practice. Futures-building exercises, such as scenario planning and speculative design workshops, offer structured ways to surface divergent assumptions about AI trajectories and their governance implications, particularly among participants who may not share a common technical vocabulary. These formats have the additional advantage of making power dynamics explicit, creating space to contest dominant framings rather than simply respond to them. Co-creation methodologies — including collaborative policy prototyping and participatory drafting sessions — allow diverse stakeholders to move from critique to proposition, producing concrete governance inputs rather than position statements. When designed well, these formats generate outputs that carry genuine multi-stakeholder ownership, which strengthens both the quality and the legitimacy of subsequent negotiations. Game-based and simulation approaches deserve particular consideration for their capacity to engage participants who are less familiar with formal multilateral settings, including community representatives, young people, and grassroots organisations. Governance simulations and role-playing exercises can build the institutional literacy needed for meaningful participation whilst simultaneously surfacing perspectives that more conventional formats would miss. Taken together, these methodologies reflect a core principle: that the quality of AI governance depends not only on what is decided, but on how decisions are reached. Innovative engagement formats are not a communications strategy — they are a governance instrument.

One of the most instructive recent examples of effective regional AI governance is the model developed around LatamGPT. Beyond its technical ambition, the initiative is notable for its process: a genuine multi-stakeholder coordination effort involving governments, academic institutions, civil society organisations, and technical communities across Latin America and the Caribbean, each functioning as active agents rather than passive recipients of a pre-defined framework. This demonstrates that inclusive, regionally-owned governance processes are not only more equitable but more contextually appropriate and politically durable. The LatamGPT experience offers a concrete, replicable model for how the AI Dialogue might structure its own participatory architecture.