Ramona AI

Success means one thing: that this Dialogue produces governance frameworks that protect people who have never been consulted in any AI policy room.Concrete outcomes that would mark a genuine success:First, a shared understanding that AI governance must be grounded in evidence from communities at risk, not only from technology developers or regulators. The voices of workers who lost jobs to fraudulent recruitment, migrants deceived by fake labor offers, and survivors of exploitation must inform what "safe and trustworthy AI" means in practice.Second, actionable recommendations on how AI systems deployed directly to vulnerable populations via accessible channels such as WhatsApp and Telegram should be governed, since most current frameworks assume internet access, digital literacy, and institutional intermediaries that do not exist for billions of people in the Global South.Third, a commitment to close the governance gap between AI systems designed in high-income countries and the realities of communities in Latin America, Africa, and Asia where these systems are deployed but where affected communities have no representation in policy processes.Fourth, recognition that AI can be a tool for protection, not only a source of risk. Ramona AI has analyzed more than 127,000 job offers across five countries with 98% accuracy, reaching workers directly on their phones in real time. This model, built on 200 interviews with survivors of labor trafficking and exploitation, demonstrates that AI governance must include protection-oriented innovation from the Global South as a reference point, not an afterthought.Success is not a document. Success is a process that makes the next generation of AI governance irreversible in its commitment to human rights, inclusion, and the protection of those most exposed to AI harms.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI

These four priorities reflect the operational reality of Ramona AI, the first AI system in the world built specifically to protect vulnerable populations from labor fraud and exploitation, accessible via WhatsApp and Telegram, operating in five countries across Latin America. Social, economic and ethical implications of AI is the foundational priority because AI does not arrive neutrally. In labor markets across the Global South, AI-powered job platforms are already being used to deceive workers, particularly young women, migrants, and people in economic precarity. Governance frameworks must grapple with these harms as a starting point, not a footnote. Protection and promotion of human rights is urgent because the populations most exposed to AI harms are the least represented in governance discussions. Ramona AI was built on 200 interviews with survivors of labor trafficking and exploitation. Their experiences revealed that fraudulent job offers follow predictable patterns that AI can detect before harm occurs. A rights-based approach to AI governance must include the right not to be deceived, recruited into exploitation, or trafficked through digital channels. Transparency and accountability matters because AI systems that affect vulnerable people must be explainable to those people, not only to regulators. Workers receiving an alert through WhatsApp need to understand why a job offer is flagged. Governance frameworks must mandate explainability at the point of impact, in the language and channel of the affected person. Safe and trustworthy AI requires expanding the definition of safety beyond cybersecurity and model alignment to include protection from social harms such as labor fraud, deceptive recruitment, and digital exploitation. An AI system is not safe if it is technically robust but operates in a governance vacuum that allows it to be weaponized against workers.

First: AI-enabled labor exploitation and fraudulent recruitment as a governance priority. Current AI governance frameworks focus on algorithmic discrimination, deepfakes, and autonomous weapons. They largely ignore how AI is already being used to scale labor fraud and recruitment into trafficking. Fraudulent job offers generated or amplified by AI reach millions of workers in the Global South each year through digital platforms, social media, and messaging apps. This is not an emerging risk; it is a present harm. The International Labour Organization estimates that 27.6 million people live in forced labor globally, with migrant workers facing three times higher risk, often recruited through deceptive offers. AI governance must include specific protections against AI-enabled deceptive recruitment. Second: Governance of AI deployed through informal digital channels. Most AI governance discussions assume that AI reaches users through regulated platforms with terms of service, appeals mechanisms, and institutional accountability. However, the most impactful AI tools for vulnerable populations often operate through WhatsApp, Telegram, and SMS, channels with no built-in governance structures. The Dialogue must develop frameworks for AI systems operating in these environments, including standards for transparency, user protection, and accountability that do not assume institutional intermediaries. Third: The paradigm shift from rescue to prevention in AI and human rights protection. Existing international frameworks for protecting vulnerable workers focus on rescue and prosecution after exploitation occurs. AI makes prevention possible at scale, through real-time alert systems that reach workers before harm, trained on evidence from survivors. The Dialogue should recognize prevention-oriented AI as a distinct and urgent category requiring dedicated governance attention, investment frameworks, and South South cooperation mechanisms that support organizations already delivering this protection in the Global South.

Latin America faces a paradox: the region is simultaneously one of the fastest adopters of AI-powered digital platforms and one of the least represented in the rooms where AI governance is decided. This gap is not abstract. It has a body count. In Mexico, Colombia, Argentina, Peru, and Chile, where Ramona AI operates, fraudulent job offers have proliferated exponentially through WhatsApp, Instagram, and Telegram. These offers target young women, internal migrants, and people in economic precarity with promises of legitimate employment that lead to exploitation, debt bondage, and in the most severe cases, labor trafficking. AI-generated content has lowered the cost of producing convincing fraudulent offers to near zero. Governance frameworks have not kept pace. The most significant challenge is that no binding international standard currently defines what constitutes a safe AI system in the context of labor recruitment. The ILO Fair Recruitment Principles exist but predate AI-powered deception at scale. National regulators in the region lack both the technical capacity and the jurisdictional reach to govern digital recruitment platforms that operate across borders with no physical presence. The most significant opportunity is that detection is now possible before harm occurs. Ramona AI has analyzed more than 127,000 job offers across five countries with 98% accuracy, identifying fraud patterns trained on 200 survivor interviews and reaching workers directly on their phones in real time, at no cost. This demonstrates that AI governance is not only about restricting harmful systems. It is also about enabling protective systems to operate, scale, and be recognized as legitimate infrastructure for human rights protection. The governance gap that most urgently needs to close is the absence of any multilateral framework that incentivizes, funds, and sets standards for prevention-oriented AI built in and for the Global South.

The AI Dialogue can do what no previous governance process has done: make the Global South a producer of AI governance frameworks, not only a recipient of frameworks designed elsewhere.Existing AI governance initiatives, from the OECD AI Principles to the EU AI Act to the AI Safety Summits, were built primarily by and for high-income countries with mature regulatory institutions, technical capacity, and direct influence over major AI developers. They have produced valuable frameworks. They have not produced frameworks that address the specific risks facing workers in Lagos, Lima, or Lahore who encounter AI-powered deception through a WhatsApp message with no institutional protection available to them.The AI Dialogue, co-chaired by El Salvador and Estonia, has a structural opportunity to change this. El Salvador's leadership signals that the priorities of Latin America and the Global South are not secondary agenda items but co-equal starting points. This must translate into concrete cooperation mechanisms.Specifically, the Dialogue can advance international cooperation in three ways. First, by establishing South-South knowledge exchange platforms where organizations building protection-oriented AI in developing countries can share models, datasets, and governance lessons without depending on North-to-South technology transfer paradigms. Second, by creating a multilateral recognition framework for AI systems that meet human rights and safety standards, so that innovators in the Global South can demonstrate compliance without navigating dozens of incompatible national regulations. Third, by connecting the outputs of this Dialogue directly to the financing mechanisms of multilateral development banks, so that governance recommendations generate actual investment in the AI infrastructure and protective systems that vulnerable communities need.The Dialogue's added value is legitimacy. One hundred ninety-three member states at the table means that agreements reached here carry the moral weight that no private or regional initiative can claim.

Several existing initiatives have built critical groundwork that the AI Dialogue must connect with rather than duplicate. The ILO Fair Recruitment Initiative has established international standards against deceptive and abusive recruitment practices affecting migrant workers. The AI Dialogue should explicitly extend these standards to cover AI-powered recruitment platforms and fraudulent job offers distributed through digital channels, closing a gap the ILO frameworks predate. The UNODC Global Action against Trafficking in Persons and Smuggling of Migrants has documented how digital platforms accelerate recruitment into exploitation. The Dialogue should formalize the connection between AI governance and anti-trafficking frameworks, recognizing that prevention-oriented AI tools are legitimate and necessary components of the international response. UNESCO Women for Ethical AI has built a network of researchers and innovators committed to gender-responsive AI development. The Dialogue should treat this network as a standing civil society advisory mechanism rather than a parallel process, integrating its recommendations into the governance architecture being built. The Global Digital Compact established the political foundation. The AI Dialogue adds operational specificity: where the GDC set principles, the Dialogue must produce actionable recommendations that translate those principles into governance standards with implementation pathways for developing countries. The UNIDO ONE World Sustainability Awards and the CAF Women in Innovation Prize have already identified innovators from the Global South producing AI solutions to social problems at scale. The Dialogue should build a formal mechanism to bring these practitioners into policy conversations, because the most important governance insights often come from organizations that have already deployed AI in contexts where failure means real harm to real people. The added value the AI Dialogue brings is integration. These initiatives exist in parallel silos. The Dialogue is the only process with the universal mandate and the multilateral legitimacy to connect them into a coherent global framework.

The AI Dialogue will succeed or fail based on whether it creates genuine exchange between people who design AI systems and people whose lives are shaped by them. Format and structure must be built around that gap.For governments, the contribution is political commitment and regulatory alignment. But governments must arrive having consulted the communities most affected by AI in their jurisdictions, not only their technology ministries. The Dialogue should require participating states to demonstrate multistakeholder consultation at the national level as a condition of meaningful engagement, not a courtesy.For civil society, the contribution is ground truth. Organizations working directly with vulnerable populations, workers, migrants, survivors of exploitation, and communities with limited digital access hold evidence that no dataset or technical report captures. Their participation must go beyond token interventions in plenary sessions. Civil society practitioners should co-chair thematic breakout sessions, contribute case evidence to the Scientific Panel, and have structured access to the summary drafting process.For the private sector, the contribution is technical transparency. Companies developing and deploying AI systems should be required to disclose, in formats accessible to non-technical stakeholders, how their systems perform across different geographies, languages, and socioeconomic contexts. Voluntary commitments made in previous summits have proven insufficient. The Dialogue should establish a baseline disclosure standard.For academia and the technical community, the contribution is independent evaluation. Research on AI systems operating in the Global South is chronically underfunded. The Dialogue should connect directly with the Scientific Panel to commission evidence reviews on AI harms and protections in developing country contexts.Structurally, the Dialogue needs fewer high-level speeches and more structured deliberation. Small working groups organized by specific governance problem, not by stakeholder category, will produce more actionable outcomes than parallel tracks that never intersect.

The most consequential voices missing from global AI governance discussions are not absent because they have nothing to say. They are absent because the architecture of international policy processes was not built for them. Workers in informal economies across Latin America, Sub-Saharan Africa, and South and Southeast Asia are among the populations most exposed to AI-enabled harms, including fraudulent recruitment, algorithmic wage suppression, and labor trafficking facilitated through digital platforms. They have no seat at any governance table. Including them requires more than translation: it requires shifting where consultations happen, through which channels, and in which languages. Survivors of labor exploitation and trafficking have direct, irreplaceable knowledge of how deceptive recruitment operates, what signals indicate fraud, and what interventions actually protect people before harm occurs. Ramona AI was built on 200 interviews with survivors across Latin America. That evidence base is more precise and more actionable than any regulatory impact assessment produced without survivor input. The Dialogue should create a formal mechanism for survivor testimony to enter the governance record. Women in rural and peri-urban areas of the Global South face compounded exposure: they are disproportionately targeted by fraudulent job offers, underrepresented in AI development, and largely absent from national AI policy processes. UNESCO Women for Ethical AI has begun to address this gap in the research community. The Dialogue must extend this to the policy community. Small and medium organizations building AI for social protection in developing countries are excluded from governance processes dominated by large technology companies and well-resourced NGOs from high-income countries. A dedicated track for Global South practitioners, with subsidized participation, interpretation support, and direct access to Co-Chairs, would begin to correct this. Inclusion is not a side event. It is the core test of whether this Dialogue produces governance that works for everyone.

The formats that have dominated international AI governance gatherings, keynote speeches, ministerial panels, side events with predetermined conclusions, have produced polished documents and limited change. The AI Dialogue in Geneva needs formats that generate friction, evidence, and genuine deliberation. Three formats would make a material difference. First, evidence walls anchored in practitioner testimony. Before each thematic breakout session, a curated set of case studies from organizations operating AI systems in developing country contexts should be displayed and discussed, not as inspiring stories but as governance problems requiring specific solutions. What worked, what failed, what policy environment enabled or obstructed the intervention. This grounds abstract governance principles in operational reality. Second, structured problem inversion sessions. Instead of asking what AI governance should look like in theory, ask participants to diagnose specific governance failures: Why did this fraudulent recruitment network operate undetected for two years across three countries? What framework would have enabled detection and intervention? Working backward from documented harms produces more precise governance recommendations than working forward from principles. Third, cross-constituency working groups with binding drafting rights. The summary documents of the AI Dialogue will be written by someone. If that process happens only among diplomats and secretariat staff, the voices of civil society practitioners and affected communities will be filtered out at the final stage. Working groups with explicit representation from civil society, academia, and Global South practitioners should have a formal role in drafting the thematic conclusions of each breakout session, not only commenting on drafts produced without them. The Dialogue of Dialogues format planned for Day 2 is promising. Its success depends on whether the dialogues being connected are genuinely multivocal or whether they simply aggregate the same dominant perspectives through different institutional containers.

Effective AI governance is not only what regulators prohibit. It is also what practitioners build. The most instructive examples combine technical rigor, human rights grounding, and operational deployment in contexts where governance gaps cause real harm. Ramona AI (Mexico, operating in five countries across Latin America) demonstrates that prevention-oriented AI is achievable at scale with limited resources. Built on 200 interviews with survivors of labor trafficking and exploitation, the platform analyzes job offers in real time through WhatsApp and Telegram, reaching workers directly on their phones at no cost, with 98% accuracy across more than 127,000 offers analyzed. Its governance model is built into its architecture: survivor testimony informs training data, no personal data is retained, alerts are explainable to the user in plain language, and the system operates under continuous human oversight. Recognized by UNIDO as a Top 5 Global Innovator and awarded the CAF Women in Innovation Prize, Ramona AI offers a replicable model for protection-oriented AI governance from the Global South. The ILO Fair Recruitment Initiative provides the closest existing policy framework for governing AI in labor recruitment contexts, establishing standards against deceptive practices that can be extended to digital and AI-powered channels. Estonia's e-governance infrastructure, represented through Co-Chair Ambassador Tammsaar's country, demonstrates that human-centered AI deployment with transparency, interoperability, and public accountability is achievable at national scale, including AI systems for public services with built in explainability requirements. The UNODC Global Action against Trafficking has begun integrating AI detection tools into anti-trafficking responses, establishing a precedent for multilateral recognition of technology-based protection mechanisms. The pattern across these examples is consistent: effective AI governance combines clear human rights objectives, survivor or community-informed design, transparent accountability mechanisms, and deployment channels that reach people where they actually are.