Prosight

The first Global Dialogue on AI Governance will succeed not by reaching consensus, that's too ambitious for a first convening, but by laying the groundwork that makes consensus possible later. Three outcomes would signal real progress. First, a shared lexicon. AI governance conversations stall because nations, regulators, and technologists don't mean the same things when they say "safety," "transparency," or "high-risk system." A successful dialogue produces a working glossary, not binding definitions, but enough common ground to hold productive disagreement. As someone building AI products deployed across thousands of users, I've seen how terminological drift creates gaps between what regulators intend and what builders implement. Closing that gap starts with language. Second, genuine inclusion of the Global South. AI is not a Western problem or a Western opportunity, but governance conversations have largely been led by the U.S., EU, and UK. A successful dialogue actively centers voices from regions that will be most affected by AI's economic disruption, yet have the least influence over how it's shaped. Success looks like those delegations leaving with structural seats at the table going forward, not just courtesy invitations. Third, a commitment to practitioner participation. Governance frameworks that are built without the people building AI tend to regulate for systems that no longer exist by the time rules take effect. A successful dialogue establishes formal channels for industry practitioners, not just lobbyists, to participate in ongoing governance development. Success isn't a signed treaty. It's a room that decides to keep meeting, with the right people in it.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• AI capacity-building

As a product manager building and deploying Agentic AI systems and the founder of ProSight, where I help enterprises design and implement AI governance frameworks, I operate at the intersection of building AI and governing it. These four priorities aren't abstract to me; they reflect the gaps I encounter in both roles every day. Safe, secure, and trustworthy AI is foundational. Agentic systems that reason and act autonomously on behalf of users can cause harm at speed and scale that reactive safeguards can't catch. In my governance work with enterprises, I consistently see organizations underestimate this risk until something breaks. Safety must be designed in from the start, not bolted on after deployment. AI capacity-building is where intention meets execution. Many enterprises want to engage with AI responsibly but lack the internal literacy, processes, or resources to do so. I see this gap daily at ProSight. Without deliberate investment in capacity, at the organizational and national level, governance becomes a conversation among the already-capable, and the divide widens. Social, economic, ethical, cultural, linguistic, and technical implications are not separate workstreams, they are the same challenge viewed from different angles. A technically sound AI system can still fail users when it ignores context, culture, or language. Governance frameworks that treat these dimensions as peripheral will consistently miss the people they're meant to protect. Transparency, accountability, and human oversight are the practical mechanisms that make the other three real. Without clear accountability chains and meaningful human control, especially in agentic settings, trust erodes and harms compound. These are the first things enterprises ask for, and the last things AI systems are built with. These four priorities are mutually reinforcing conditions for AI that genuinely serves people and organizations.

The themes identified in Resolution 79/325 provide a strong foundation, but several cross-cutting issues sit outside their boundaries, issues that are already shaping how AI is built, deployed, and experienced in practice. Agentic AI and autonomous decision-making deserve dedicated attention. Current governance frameworks were largely designed for AI systems that respond to inputs. Agentic systems, those that plan, act, and chain decisions across tools and workflows with minimal human intervention, introduce a fundamentally different risk profile. Questions of accountability, reversibility, and meaningful oversight become significantly more complex when AI is acting, not just answering. AI and labor and economic displacement remains underrepresented in governance conversations relative to its scale. AI is already reshaping knowledge work across industries, and the pace is accelerating. Without explicit attention to workforce transition, income redistribution, and the social contracts that hold economies together, governance frameworks risk addressing technical risk while missing human consequence. Governance inequality between the Global South and wealthier nations is a structural issue that cuts across every listed theme. In my work with enterprises, I observe that organizations in under-resourced regions often inherit governance frameworks designed elsewhere, for contexts that don't match their own. The Dialogue should treat this asymmetry as its own agenda item, not a footnote. AI identity and synthetic media risks, including deepfakes, synthetic voices, and AI-generated personas, are evolving faster than existing legal and ethical frameworks can accommodate. The implications for truth, trust, and democratic participation are significant and not fully captured under existing transparency or safety themes. These issues are not future concerns. They are present realities that merit explicit space in the Dialogue.

Living in the United States as a Mexican immigrant gives me an unusual vantage point: I am inside the world's most advanced AI ecosystem, yet I carry the lived experience of a region it rarely designs for. In the US, the governance gaps I feel most acutely are not about access, they are about accountability. As someone building Agentic AI systems and helping enterprises implement governance frameworks, I see organizations move fast and govern slowly. Safety and transparency are treated as compliance checkboxes rather than design principles. Human oversight is often nominal, present on paper, absent in practice. The opportunity here is real: the US has the talent, infrastructure, and institutional capacity to lead. The challenge is the will to slow down enough to do it responsibly. But when I think of Mexico and Latin America, the picture shifts entirely. The governance gaps are not just structural, they are felt. AI systems trained predominantly on English data and Western cultural norms are being deployed in Spanish-speaking, multilingual, and indigenous communities without meaningful adaptation. The economic displacement risk is acute in a region where large portions of the workforce are in roles AI will restructure first. And yet Latin America is largely absent from the rooms where governance decisions are made. The thematic areas I selected, safety, capacity-building, social and cultural implications, and transparency, map directly onto this contrast. In the US, the gaps are about enforcement and accountability. In Latin America, they are about inclusion and survival. I carry both realities. That is precisely why I believe governance conversations must stop treating these as separate problems.

One of the most striking things I observe in my work, building AI products and helping enterprises implement governance frameworks, is how fragmented the global conversation is. Nations are legislating in parallel, companies are self-regulating inconsistently, and practitioners like me are navigating a patchwork of standards that often contradict each other. The AI Dialogue can be the connective tissue this moment needs. Norm-setting and shared standards are where international cooperation has the highest leverage. Not every country needs identical regulation, but shared baseline definitions of safety, accountability, high-risk use, would reduce the compliance chaos enterprises face today and create a more level, trustworthy global AI environment. Knowledge and capacity exchange may be the most immediately impactful role. I've worked with enterprises in Latin America that are hungry to govern AI responsibly but are starting from scratch, without the institutional knowledge that more resourced regions take for granted. The Dialogue can formalize channels for that transfer, not as charity, but as mutual investment in a safer global ecosystem. Bridging practitioners and policymakers is a gap I live in daily. Governance frameworks designed without input from people actually building AI tend to arrive late, target outdated systems, and create compliance burdens without reducing real risk. The Dialogue should establish structured, ongoing mechanisms for practitioners to inform, not just react to policy. Amplifying underrepresented voices is not optional. As someone who carries both a US and a Latin American perspective, I know how much is lost when governance is shaped by a narrow set of experiences. The Dialogue's legitimacy depends on who is in the room. International cooperation on AI governance won't happen by accident. The Dialogue is one of the few spaces positioned to make it intentional.

In my work helping enterprises implement AI governance, one of the first things I do is map the existing landscape and it is crowded. The challenge isn't a shortage of initiatives. It's that they don't talk to each other, and the organizations that need them most often don't know they exist. Several efforts have laid genuinely important groundwork. The OECD AI Principles established an early shared vocabulary for trustworthy AI that many national frameworks have since built on. The UNESCO Recommendation on the Ethics of AI brought a values-based lens and meaningful Global South participation. The EU AI Act is the most comprehensive attempt at binding regulation, offering a risk-tiered model others are watching closely. The UN Secretary-General's Advisory Body on AI produced actionable recommendations specifically aimed at international governance gaps. Regional and civil society initiatives, from the African Union's AI strategy to grassroots digital rights organizations are doing critical work that rarely surfaces in formal multilateral settings. The AI Dialogue's added value is not in replacing any of these. It is in connecting them. Each existing initiative operates within its own lane, geographic, thematic, or institutional. None has the mandate or the convening power to create coherence across all of them. What the Dialogue can uniquely offer is a synthesis function: translating OECD principles into capacity-building programs for under-resourced nations, connecting EU regulatory frameworks with practitioner realities, and amplifying regional voices that multilateral processes have historically sidelined. As someone navigating this landscape daily, both as a builder and a governance advisor, I can say clearly: the infrastructure exists. What's missing is the architecture that links it.

Every governance forum I've participated in, whether inside enterprises or at the policy level, shares a common blind spot: the people most affected by AI decisions are the least represented in the rooms where those decisions are made. The AI Dialogue has an opportunity to break that pattern, but only if it's designed to. Civil society and affected communities should not be afterthoughts in the Dialogue's structure. They should be co-designers of it. This means dedicated seats, not open-mic moments, for community organizations, labor advocates, indigenous groups, disability rights voices, and representatives from regions bearing the greatest risk of AI-driven displacement. Their contribution is not anecdotal. It is ground truth that no technical or regulatory stakeholder can replicate. Governments and regulators bring institutional authority and implementation power. Industry and practitioners bring technical literacy and operational reality. Academia brings longitudinal research and independent analysis. Each group is necessary. None is sufficient alone. The Dialogue's structure should reflect this interdependence, not through siloed tracks, but through deliberately mixed working groups where these perspectives are in active dialogue with each other. On format, I'd recommend three structural principles. First, persistent working groups between convenings, a single annual meeting produces declarations, not change. Second, transparent documentation that is publicly accessible, translated, and usable by organizations without legal or policy teams. Third, a formal feedback loop so that communities and practitioners can see how their input shaped outcomes, not just that it was received. The Dialogue's legitimacy will ultimately be measured not by who attended, but by whose realities shaped what was decided.

I'll answer this question personally, because I am one of the underrepresented voices im asking about. As a Mexican woman working at the frontier of AI, building agentic systems, advising enterprises on governance, and navigating spaces where I am frequently the only person who looks like me, I experience the representation gap firsthand. It shapes what problems get prioritized, what risks get named, and whose futures get protected. Latin America and the Global South are largely absent from AI governance conversations in any structural way. This is not for lack of expertise or perspective, it is a function of who gets invited, who can afford to participate, and whose frameworks get treated as universal. AI systems are being deployed across Latin America at scale, in Spanish, Portuguese, and dozens of indigenous languages, often without the cultural or linguistic grounding to serve those communities well. The people most affected are generating the least governance input. Women and gender minorities in AI face a compounding invisibility. We are underrepresented among builders, underrepresented in policy rooms, and disproportionately affected by AI systems that encode historical bias in hiring, lending, healthcare, and safety. Our absence from governance conversations is not incidental, it produces frameworks with predictable blind spots. Inclusion requires more than invitations. It requires removing the structural barriers that make participation inaccessible: translation and interpretation as a baseline, funded participation for representatives from under-resourced regions, asynchronous contribution mechanisms for those who cannot travel, and governance processes that recognize oral traditions and community knowledge as valid alongside academic credentials. Representation isn't diversity theater. It is the difference between governance that works on paper and governance that works in practice.

In product work, we've learned that the format of a conversation determines the quality of its output. A room full of experts delivering prepared remarks produces polished consensus. It rarely produces honest reckoning. The AI Dialogue needs formats designed for the latter. Scenario-based working sessions are among the most effective formats I've used with enterprise teams. Rather than debating abstract principles, participants work through concrete AI deployment scenarios, a hiring algorithm in Brazil, an agentic system making financial decisions in rural Kenya, and surface governance gaps through specificity. Problems that feel distant in theory become urgent when they're named. Red team panels, where a small group is explicitly tasked with challenging the prevailing consensus, create the productive friction that plenary sessions avoid. In product development, structured adversarial review consistently surfaces blind spots that collaborative sessions miss. Governance conversations need the same discipline. Asynchronous deliberation tracks running between in-person convenings would significantly expand who can meaningfully participate. Not everyone can attend a convening in Geneva or New York. Digital-first contribution mechanisms, structured input forms, moderated async forums, translated discussion threads, allow communities without travel budgets or institutional backing to shape outcomes, not just observe them. Practitioner clinics, small, focused sessions where policymakers engage directly with builders, operators, and affected communities on specific implementation challenges, bridge the gap I see constantly in governance work: rules written by people who've never deployed an AI system, for systems they've never seen in production. Format is not logistics. It is a statement about whose knowledge counts and how decisions get made. The Dialogue should design its formats with the same intentionality it brings to its agenda.

After years of building AI products and helping enterprises govern them, I've come to believe that effective AI governance shares one trait: it is specific enough to be actionable and flexible enough to survive contact with reality. These are the examples I return to most often. The EU AI Act remains the most instructive policy framework, not because it's perfect, but because it introduced risk-tiering as a governance logic. Treating a hiring algorithm and a spam filter as categorically different regulatory problems is the right instinct. Enterprises I work with use this tiering model even outside EU jurisdiction because it gives governance conversations a practical entry point. NIST's AI Risk Management Framework is the most useful practitioner-facing policy tool I've encountered. It is non-prescriptive enough to adapt to different organizational contexts while structured enough to create accountability. At ProSight, it anchors how we help enterprises build internal AI governance programs from the ground up. Model cards and datasheets for datasets, technical documentation practices pioneered in research, have quietly become one of the most effective transparency tools available. When enterprises I advise adopt them, it forces specificity about what a system was trained on, what it performs well at, and where it fails. That specificity is where accountability begins. Brazil's AI regulatory process deserves more international attention. Its multi-stakeholder development approach, actively incorporating civil society, academia, and industry across multiple public consultation rounds, offers a replicable model for regions building governance capacity without simply importing frameworks designed elsewhere. What these examples share is that they treat governance as infrastructure, not compliance theater. That distinction is everything.