Veilleurs du Web at UNICEF en RDC

For the first Global Dialogue on AI Governance to be a success, it must deliver concrete outcomes regarding the protection of the data that AI consumes. First, a binding commitment to data provenance and consent: all training data, especially personal or copyrighted material, must be traceable and lawful, with effective opt-out mechanisms. Second, the adoption of privacy by design as a non-negotiable standard, using techniques such as differential privacy or federated learning, backed by dedicated funding for privacy‑enhancing technologies. The Dialogue must also address data colonialism: vulnerable populations in the Global South must retain control over their data, and a code of conduct should govern its extraction and ensure fair benefit‑sharing. Finally, a roadmap for cross‑border redress is essential: any individual must be able to challenge abusive use of their data, regardless of where the AI company is based. Without these advances, AI governance would remain incomplete and risk a race to the bottom. Success will therefore be measured by the adoption of these operational principles, not by mere declarations of intent.

• Safe, secure and trustworthy AI
• Open-source software, open data and open AI models
• Protection and promotion of human rights
• Interoperability of governance approaches

My entity prioritizes safe, secure and trustworthy AI because without foundational safety guarantees and robustness against misuse, all other benefits of AI become secondary. Trust is the prerequisite for adoption and sustainable governance. Interoperability of governance approaches is urgent to avoid fragmentation. Divergent national regulations will increase compliance costs, create loopholes, and hinder cross-border cooperation. We need common technical standards and mutual recognition frameworks to enable seamless AI innovation while protecting public values. Protection and promotion of human rights must remain the non-negotiable core of AI governance. AI systems can amplify discrimination, erode privacy, and undermine due process. Active engagement is required to embed human rights impact assessments, redress mechanisms, and safeguards against automated decision-making that affects people's lives. Open-source software, open data and open AI models are priorities because transparency drives accountability. Open ecosystems enable independent auditing, democratize access to AI tools for lower-resource actors, and prevent concentration of power among a few corporations. However, openness must be balanced with responsible release protocols to mitigate risks of malicious use

Yes, several cross-cutting and emerging issues are not fully captured by the listed themes. Environmental sustainability is a pressing omission. The carbon footprint of training and deploying large AI models, as well as water consumption for data center cooling, raises urgent climate concerns. Governance must include efficiency standards, disclosure requirements, and incentives for green AI. Labor and the future of work goes beyond general "social implications." Specific attention is needed on automation-induced displacement, working conditions for data labelers and content moderators, and collective bargaining rights in AI-augmented workplaces. These labor dimensions are often overlooked in high-level governance discussions.

The governance gaps shape the challenges and opportunities for fact-checking journalists in eastern DRC. Significant Challenges : The absence of interoperability and enforceable human rights protections allows disinformation to be weaponised by armed groups to incite violence, exacerbating the region's conflict. Fact-checkers like Eleza Fact operate in a climate of fear, facing harassment, arrests, and even killings while their access to information is blocked by armed groups and the government. Furthermore, a lack of governance around safe, secure, and trustworthy AI means the same generative AI tools meant to empower can easily be used against them. Malicious actors already leverage AI-generated deepfakes and "bot-like" accounts to manipulate public opinion and target civil society, making it harder for communities to distinguish real from synthetic content. Opportunities through Open-Source and AI Governance : Conversely, open-source AI models offer powerful tools for fact-checking. Eleza Fact uses natural language processing to detect hate speech on community radios, automatically generating counter-narratives. With better interoperability and governance, these open tools could be shared and scaled across the region safely. Finally, closing the governance gap on human rights can turn AI into a shield. By embedding trauma-informed AI to protect journalists and establishing international standards for critical infrastructure, vulnerable communities can regain access to trusted information. Good governance would transform AI from a weapon of disinformation into a tool for truth

The AI Dialogue can advance international cooperation on AI governance by fulfilling four critical roles, directly addressing the gaps and opportunities previously identified. Norm-setting through inclusivity: The Dialogue provides a legitimate, UN-anchored platform to develop shared principles on data protection, human rights, and trustworthy AI. By bringing together governments, civil society, and tech companies from the Global North and South, it can produce non‑binding frameworks that later harden into customary norms, especially on issues like synthetic content or biometric surveillance. Bridging interoperability gaps: Fragmented regulations (EU's AI Act, US executive orders, China's provisions) increase compliance costs and create loopholes. The Dialogue can map existing approaches, identify common technical standards (e.g., data provenance, privacy‑preserving audits), and propose mutual recognition mechanisms. This reduces friction while maintaining safeguards. Capacity‑building for the Global South: Eastern DRC's fact‑checkers cannot benefit from open‑source AI without computing power, skills, and legal protection. The Dialogue can launch a voluntary AI Capacity Fund, turning rhetoric into resources, to support underserved regions in developing governance infrastructure, training fact‑checkers, and securing access to safe AI tools. Accelerating response to emerging threats: AI's military, disinformation, and labor impacts evolve faster than treaties. The Dialogue can act as an early‑warning system and rapid‑response coordination hub, enabling joint actions such as shared deepfake detection repositories or emergency protocols for AI‑amplified violence

The Global Dialogue on AI can build upon several existing initiatives: the OECD AI Principles and GPAI (ethical frameworks), the Council of Europe Framework Convention (legally binding treaty on human rights), the UNESCO Recommendation (ethical norms adopted by 193 states), the AI for Good Summit (development applications), the Independent International Scientific Panel on AI (evidence base), and the African Union's Declaration on AI. The Dialogue's added value lies in four contributions. First, strategic coordination: it aligns these fragmented initiatives, avoids duplication, and promotes interoperability. Second, global inclusivity: with UN legitimacy, it brings in Global South countries, including fragile states like the DRC, often absent from closed clubs like GPAI. Third, it bridges norms and operational action: translating high-level principles (OECD, UNESCO) into practical guidance for local sectors (fact‑checking in Kivu) and cross‑border redress mechanisms. Fourth, it connects AI governance with peace and security: explicitly addressing deepfakes and disinformation in conflict zones, where other initiatives have remained vague. In doing so, the Dialogue transforms a scattered landscape into a coherent, equitable, and operational ecosystem.

• Different stakeholders can each play a distinct role in the AI Dialogue. Governments should share regulatory experiences and commit to interoperability. Civil society, including journalists, must bring frontline evidence of AI harms. Private sector actors should disclose model data, safety protocols, and collaborate on red-teaming. Academia provides independent audits and benchmarks, while International Organisations offer existing frameworks. For format and structure, I recommend a hybrid, multi-track model: plenary sessions (virtual plus regional hubs) for high-level commitments
• permanent working groups on data protection, synthetic content, capacity-building, and conflict-sensitive AI, delivering actionable outputs
• regional dialogue forums hosted by local partners to ensure context-specific input from fragile states
• a multistakeholder advisory body to co-design agendas
• open calls for written contributions with public consultations
• and a transparent reporting dashboard tracking commitments. The Dialogue should avoid closed-door sessions, prioritize pre-scheduled briefings for underrepresented groups (youth, indigenous communities, conflict-affected journalists), and provide funding for their participation (travel, connectivity, interpretation). A rotating co-chair model between Global North and South would build trust and ownership. This structure ensures inclusivity, accountability, and actionable outcomes.

• Currently underrepresented voices in global AI governance include conflict-affected communities (e.g., eastern DRC), indigenous peoples, rural populations, informal workers, women and gender minorities in low-resource settings, people with disabilities, small island states, and frontline fact-checkers who face AI-generated disinformation daily. These groups often lack the technical literacy, connectivity, funding, or political leverage to attend closed-door forums. To include them meaningfully, the AI Dialogue should: establish regional listening hubs with interpretation and facilitation in local languages
• provide dedicated funding for grassroots organizations to participate (travel, childcare, assistive technology)
• create a standing civil society advisory council with reserved seats for underrepresented constituencies
• mandate video and written submissions that are summarized and integrated into official agendas
• and deploy mobile and offline consultation tools (SMS, radio call-ins) for communities without reliable internet. Additionally, the Dialogue should adopt accessible formats (plain language, sign language, audio descriptions) and enforce a rotating co-chair principle ensuring Global South leadership. Without these structural changes, AI governance will remain a conversation among the few, producing rules that ignore the many.

To foster meaningful and dynamic engagement, the AI Dialogue should adopt several innovative formats. Scenario‑based policy prototyping labs would bring diverse stakeholders together to simulate real‑world AI crises (e.g., a deepfake‑fueled ethnic conflict) and co‑design rapid response mechanisms, turning abstract principles into actionable muscle memory. Reverse panel sessions flip the script: frontline fact‑checkers, indigenous data custodians, and workers from conflict zones ask questions directly to policymakers and tech executives, with no pre‑screened questions. AI‑augmented deliberation could use real‑time translation, sentiment clustering, and interactive polling to surface consensus and dissent from both in‑room and remote participants, displayed on a live "heat map of concerns." Virtual reality immersion corridors would allow delegates to experience, for example, the information environment of a rural community in the DRC or a coastal village threatened by climate‑AI models, building empathy before debate. Offline‑first engagement via community radio call‑ins and SMS voting ensures participation where connectivity is scarce, with summaries played aloud during plenaries. Finally, open space technology, self‑organised breakout sessions on emergent topics, documented and fed back into working groups, would capture grassroots ingenuity. These formats move beyond passive speeches toward iterative, inclusive, and actionable co‑creation.

Several concrete policies and practices offer effective AI governance models. The OECD AI Principles and EU AI Act (risk-based classification with binding rules) provide regulatory guardrails, while Texas's TRAIGA introduces a sandbox and social scoring ban. Technical standards like ISO/IEC 42001 (AI management systems) and NIST AI Risk Management Framework guide implementation. The C2PA standard enables content provenance to fight disinformation. On compute governance, the EQUAL Compute Network subsidizes cloud credits and infrastructure for the Global South. Multi-stakeholder platforms such as the IGF Policy Network on AI and the International Network of AI Safety Institutes coordinate shared testing protocols and interoperability. These instruments, ranging from risk-based regulation and open technical standards to subsidized compute and collaborative networks, provide a rich toolkit that the UN AI Dialogue can build upon to advance global governance.