Independent Researcher

The first Global Dialogue on AI Governance will be truly successful if it produces outcomes that go beyond declaratory consensus and translate into concrete, enforceable, and inclusive governance commitments. From a civil society perspective rooted in Latin America, success requires, first and foremost, meaningful participation from the Global South — not as recipients of AI governance frameworks designed elsewhere, but as co-architects of standards that reflect diverse legal traditions, institutional realities, and development needs. A successful Dialogue should establish shared baseline principles for algorithmic transparency and accountability in public administration, recognising that governments worldwide are deploying automated decision-making systems that affect fundamental rights — in areas such as social benefits, tax enforcement, migration, and criminal justice — often without adequate oversight mechanisms. The Dialogue should also produce a common framework for human rights impact assessments of AI systems, bridging the gap between existing instruments such as the UN Guiding Principles on Business and Human Rights and the operational realities of algorithmic governance. Additionally, success requires acknowledging AI as a vector of corruption risk. Opaque algorithmic systems in public procurement, revenue administration, and regulatory enforcement can entrench or amplify corruption. The Dialogue should explicitly integrate anti-corruption and integrity considerations into AI governance frameworks, building on existing mechanisms such as UNCAC and MESICIC. Finally, the Dialogue must produce capacity-building commitments with adequate financing, particularly for developing countries, ensuring that governance frameworks are not only adopted on paper but implemented in practice. In sum: success means a Dialogue that is not a talking shop, but a foundation for a legitimate, rights-based, and genuinely multilateral AI governance architecture.

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

These priorities are deeply interconnected and reflect the most urgent governance challenges observed from a Latin American civil society perspective. Transparency, accountability, and human oversight is foundational. Across the region, governments are deploying automated decision-making systems in tax administration, social benefit allocation, and law enforcement - often without disclosure of the underlying logic, meaningful appeal mechanisms, or independent audit capacity. Without enforceable transparency and oversight standards, all other governance commitments remain aspirational. This priority must include not only private sector AI but, critically, AI systems operated by public authorities, where power asymmetries with affected individuals are most acute. Protection and promotion of human rights must serve as the non-negotiable normative anchor of the entire AI governance architecture. AI systems are already producing documented harms - discriminatory outputs, erosion of due process, surveillance of civil society, and suppression of dissent. Regional human rights bodies, including the Inter-American Commission on Human Rights, have begun engaging with these issues, but binding regional standards remain absent. The AI Dialogue must produce clear guidance on how existing human rights obligations apply to AI systems, and establish accountability mechanisms for violations. Social, economic, ethical, cultural, linguistic and technical implications of AI acknowledges that AI governance cannot be reduced to technical standard-setting. The deployment of AI systems carries profound implications for labour markets, democratic participation, cultural diversity, and epistemic justice - particularly in developing countries, where governance frameworks are often imported without adequate adaptation to local legal traditions, languages, and institutional realities. Any credible multilateral AI governance framework must centre the voices and priorities of the Global South, not treat them as afterthoughts. Together, these priorities form a coherent agenda: rights-based, transparency-grounded, and globally inclusive.

One critical cross-cutting issue largely absent from the listed themes is the intersection of AI governance and anti-corruption / public integrity frameworks. As governments increasingly deploy AI systems in tax administration, public procurement, social services allocation, and law enforcement, the opacity of these systems creates novel corruption risks: algorithmic capture by private interests, discriminatory enforcement that can be weaponised, and the erosion of the right to explanation and administrative appeal. Existing international anti-corruption instruments - including UNCAC and regional mechanisms such as MESICIC - were not designed with automated decision-making in mind. The AI Dialogue presents a unique opportunity to close this gap by developing AI-specific integrity standards for the public sector. A second emerging issue is linguistic and epistemic justice in AI governance. The dominant frameworks are produced overwhelmingly in English, by actors from high-income countries, and embed assumptions about legal systems, institutional capacity, and social values that do not translate universally. Governance of AI must include the governance of whose knowledge and languages shape AI systems themselves. Third, the democratic governance of AI deserves explicit attention. AI systems are reshaping political communication, electoral processes, and public deliberation in ways that existing electoral and media regulation cannot adequately address. Finally, fiscal and revenue implications of AI - including the tax treatment of AI-generated value, data as an asset, and the role of AI in tax compliance - represent an underexplored dimension with significant equity implications for developing countries dependent on effective domestic revenue mobilisation.

Latin America and the Caribbean presents a paradox in AI governance: a region with rich constitutional traditions protecting due process, equality, and access to justice, yet with significant institutional gaps in translating those protections into effective oversight of automated systems. In Argentina and across the region, public administrations are increasingly deploying AI-assisted tools in tax enforcement, social programme eligibility, and judicial risk assessment — often through opaque procurement processes and without mandatory disclosure of algorithmic criteria. This creates a structural accountability deficit: affected individuals cannot challenge decisions they cannot understand, and oversight bodies lack the technical capacity to audit systems they did not design. The governance gap in transparency is particularly acute. Most countries in the region lack specific legislation governing automated decision-making in the public sector. Existing administrative law frameworks — designed for human decision-makers — do not adequately address the right to explanation, the auditability of training data, or the traceability of algorithmic outputs. Regional mechanisms such as MESICIC, while effective in promoting anti-corruption standards, have yet to systematically address AI as a vector of integrity risk in public administration. From a human rights perspective, the Inter-American Commission on Human Rights has begun engaging with AI-related concerns, but binding regional standards remain absent. Civil society organisations in the region operate with limited resources and technical expertise to monitor AI deployments effectively, widening the gap between formal rights and practical accountability. The opportunity, however, is real: Latin America has a generation of legal scholars, civil society actors, and public officials who understand both the technical dimensions of AI and the region's institutional realities. The AI Dialogue can catalyse regional standard-setting that is rights-based, contextually grounded, and capable of informing global frameworks — rather than merely receiving them.

The AI Dialogue occupies a unique and necessary position in the international governance landscape: it is the only truly universal forum where AI governance can be discussed with the legitimacy that only the United Nations system can provide. Its role should be distinct from — and complementary to — existing technical bodies, regional organisations, and multistakeholder initiatives. Concretely, the AI Dialogue can advance international cooperation in three ways. First, it can serve as a normative convergence platform — not by imposing a single regulatory model, but by identifying shared principles that can accommodate diverse legal traditions, development levels, and institutional capacities. The coexistence of the EU AI Act, various national strategies, and the near-absence of binding frameworks in much of the Global South creates a fragmented landscape that disadvantages countries with less regulatory capacity. The Dialogue can help bridge this gap. Second, the Dialogue can function as a mutual accountability mechanism — enabling states and stakeholders to report on implementation of AI governance commitments, share lessons learned, and identify where international support is needed. This is particularly critical for developing countries, where governance frameworks often exist on paper but lack the institutional infrastructure for effective implementation. Third, and perhaps most importantly, the Dialogue can ensure that international AI governance is not captured by the interests of a small number of technologically advanced states and corporations. By centring the voices of civil society, developing countries, and affected communities — particularly from the Global South — it can produce governance frameworks that are genuinely multilateral in character, not merely universal in name. The Dialogue's legitimacy will ultimately rest on its inclusivity and its capacity to translate deliberation into concrete, actionable, and monitored commitments

The AI Dialogue should position itself as a connector and amplifier of existing governance efforts, avoiding duplication while adding the unique value of universal legitimacy. Key initiatives it should build upon include: the OECD AI Principles and the work of the OECD.AI Policy Observatory, which offer the most developed cross-national framework for trustworthy AI, though their reach remains largely limited to high-income countries; the UNESCO Recommendation on the Ethics of AI (2021), which provides a genuinely universal normative foundation and deserves greater integration into operational governance discussions; the Global Digital Compact adopted at the Summit of the Future (2024), which established important commitments on AI governance and data that the Dialogue should operationalise; and regional mechanisms such as the Council of Europe's AI Convention, the African Union's emerging AI frameworks, and Inter-American human rights standards. From an anti-corruption and public integrity perspective, the Dialogue should also engage with UNCAC implementation review mechanisms and regional bodies such as MESICIC, recognising that AI governance in the public sector is inseparable from broader integrity frameworks. The added value of the AI Dialogue lies precisely in what no existing initiative can fully provide: a universal, intergovernmental forum with civil society participation that can produce governance outcomes carrying the normative weight of the UN system. It can serve as the space where technical standards developed by bodies like the ITU or ISO are connected to human rights obligations, development imperatives, and democratic accountability requirements. To realise this potential, the Dialogue must resist becoming merely a coordination mechanism for existing elite processes, and instead actively amplify the perspectives of underrepresented regions, languages, and communities.

The AI Dialogue's legitimacy will depend not only on who participates, but on whether participation is substantive rather than ceremonial. Different stakeholder groups bring irreplaceable contributions that the Dialogue's structure must actively accommodate. Governments bear primary responsibility for translating governance commitments into binding law and policy. Their participation should be oriented toward concrete reporting on national implementation, identification of capacity gaps, and willingness to subject AI deployments — particularly in the public sector — to international scrutiny. Civil society organisations bring ground-level knowledge of how AI systems affect communities, and serve as independent monitors of both corporate and governmental conduct. Their participation must go beyond formal consultation slots: civil society should have meaningful input into agenda-setting, access to technical working groups, and the ability to submit documented evidence of AI-related harms. The technical community and academia can bridge the gap between policy aspiration and technical reality, providing independent assessments of what governance measures are feasible and effective. Their participation should be structured to avoid capture by industry-aligned actors. The private sector has obligations, not merely interests, in AI governance. Participation should be conditioned on transparency regarding AI system deployments, supply chains, and lobbying positions. Structurally, the Dialogue should adopt a hybrid model combining plenary sessions for political commitment-making with smaller thematic working groups where substantive deliberation can occur. Intersessional processes — including written consultations, regional preparatory meetings, and open online participation — are essential to ensure that engagement is not limited to those with resources to travel to Geneva or New York. Finally, all sessions should be conducted with simultaneous interpretation in all UN official languages, and documentation made available in accessible formats, as a basic condition of genuine inclusivity.

The most significant gap in global AI governance discussions is not technical — it is representational. The voices shaping international AI governance frameworks remain overwhelmingly concentrated among actors from high-income, English-speaking countries, large technology corporations, and elite academic institutions. The most critically underrepresented voices include: communities from the Global South — particularly from Africa, Latin America and the Caribbean, and small island developing states — who are subject to AI systems they did not design, governed by frameworks they did not author, and often excluded from the forums where those frameworks are debated due to resource, language, and logistical barriers. Indigenous peoples and local communities, whose data, languages, and cultural heritage are increasingly incorporated into AI systems without free, prior, and informed consent, and whose governance traditions offer valuable alternative frameworks for thinking about collective rights and responsibilities in technological governance. Women and gender-diverse individuals, who face documented patterns of algorithmic discrimination and are underrepresented in both AI development and governance institutions. Affected communities — people subjected to AI-assisted decisions in welfare, criminal justice, migration, and employment — whose lived experience of algorithmic harm is the most relevant evidence base for governance design, yet who are almost entirely absent from international forums. Inclusion requires more than invitation. It requires dedicated funding for participation, advance provision of documents in multiple languages, regional preparatory processes that allow communities to develop positions before global meetings, and structured mechanisms — such as community hearings or testimony procedures — that give affected voices formal standing in the Dialogue's deliberative processes.

Meaningful engagement in a dialogue of this scope and complexity requires moving beyond the traditional UN format of prepared statements and negotiated texts. Structured deliberation formats — such as moderated roundtables with equal speaking time for government, civil society, technical, and affected community representatives — can produce richer exchanges than sequential statements. The OECD Global Anti-Corruption & Integrity Forum offers a useful model of thematic sessions that combine expert panels with open floor participation in a genuinely interactive format. Scenario-based working sessions can ground abstract governance debates in concrete cases. Presenting real-world examples of AI deployment in public administration — anonymised where necessary — and asking participants to diagnose governance failures and propose remedies generates more actionable outputs than general principle discussions. Pre-Dialogue regional consultations, conducted in local languages and with dedicated support for civil society participation, can ensure that the perspectives of underrepresented regions are synthesised and formally presented at the global level, rather than lost in the translation to international diplomatic language. Asynchronous and digital participation mechanisms — including open written consultation periods, online deliberation platforms, and recorded sessions with accessible transcripts — are essential to democratise participation beyond those with travel budgets and visa access. The AI Dialogue should itself model the inclusive, accessible governance it advocates. A civil society and affected communities forum, held immediately before the plenary sessions, could allow non-governmental actors to consolidate positions, identify areas of consensus, and present joint recommendations with greater collective weight. Finally, the Dialogue should commit to transparent documentation of outcomes — including minority positions and unresolved disagreements — rather than producing only consensus language that obscures the real contours of the debate. Honest documentation is itself a form of accountability.

Effective AI governance frameworks share a common architecture: clear legal obligations, independent oversight, meaningful transparency, and accessible remedy mechanisms. Several existing approaches offer valuable models. The European Union AI Act (2024) represents the most comprehensive binding regulatory framework to date, introducing risk-based classification of AI systems, mandatory conformity assessments for high-risk applications, and explicit prohibitions on certain uses. While designed for a high-capacity regulatory environment, its core architecture - particularly the concept of high-risk AI in public sector decision-making - offers transferable lessons for other jurisdictions. Brazil's draft AI regulatory framework is particularly instructive for Latin America: it explicitly addresses automated decision-making in the public sector, establishes rights to explanation and human review, and was developed through an open multi-stakeholder consultation process. It demonstrates that developing countries can design contextually appropriate governance frameworks rather than simply importing external models. At the institutional level, Argentina's experience with algorithmic tools in tax administration - including the use of risk-scoring systems by the Federal Administration of Public Revenue (AFIP/ARCA) - illustrates both the potential and the risks of AI in public administration. The absence of mandatory disclosure requirements and independent audit mechanisms in this context serves as a cautionary case, highlighting the governance gaps that urgently need to be addressed through specific legislation. The UNESCO Recommendation on the Ethics of AI (2021) offers a universally adopted normative foundation, including provisions on transparency, accountability, and human rights, that national frameworks should actively operationalise. Finally, civil society-led algorithmic auditing initiatives - such as those developed by Algorithm Watch in Europe and emerging equivalents in Latin America - demonstrate that independent technical scrutiny of AI systems is feasible and necessary, and should be formally recognised and resourced within governance frameworks.