Self

A successful first Dialogue would do three things: center people over hype, center the periphery over the usual capitals, and center enforcement over declarations. First, it should explicitly affirm that all AI governance must be grounded in international human rights law, not just voluntary "ethical principles," and that this applies equally to private platforms and states. That means concrete language on non-discrimination, freedom of expression, privacy, and the right to an effective remedy when systems cause harm. Second, it should move beyond New York (and Silicon Valley)-centric narratives by creating structured space and funding for participation from the Majority World: Nairobi, São Paulo, Kolkata, Baghdad, Tirana, and beyond. The Dialogue will be a failure if those most affected by extraction, ghost work, surveillance, and content-moderation trauma are reduced to case studies instead of agenda-setters. Third, it should produce specific, time-bound work plans for the new Scientific Panel and the Dialogue itself: e.g., commitments on interoperable safety disclosures, baseline transparency obligations for high-risk systems, and support for public infrastructure such as open models and datasets governed as digital public goods. Finally, success would look like broad agreement that AI is a global commons issue, similar to climate, with clear exclusions for military applications and a focus on sustainable development, mental health, and information integrity rather than just productivity gains.

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

From a #CovfefeTV perspective (a multilingual, global, satirical newsroom), these four priorities are tightly interlinked. "Safe, secure and trustworthy AI" must be defined from the user's side, not the vendor's. Safety includes mental-health impacts of AI companions and "therapist" chatbots, manipulation at scale through recommender systems, and the offloading of care work to unregulated systems, especially in environments like Brazilian and Kenyan WhatsApp. If governance is not anchored in the protection and promotion of human rights, safety will be reduced to infrastructure security and IP protection. Resolution 79/325 rightly emphasizes human rights, sustainable development, and non-military uses as core pillars: the Dialogue must operationalize that with guidance on high-risk use cases, red lines, and remedies. Transparency, accountability, and human oversight are essential because today's AI harms are often invisible and untraceable (moderators in Nairobi, data labelers in the Philippines, ghost workers whose labor and trauma are buried in "binary graves" of training data. Without mandatory disclosures, impact assessments, and meaningful recourse, trust is impossible. Finally, the social, economic, ethical, cultural, linguistic and technical implications of AI are where most people actually experience the technology: as job restructuring, language loss, political polarization, and new forms of dependency on foreign platforms. A global dialogue that ignores these dimensions in favor of abstract principles will miss the lived reality we report on every day.

Three cross-cutting issues deserve explicit treatment. First, mental health and psychosocial integrity. Current themes do not name the psychological impacts of AI: from parasocial dependence on chatbots, to the strain on human moderators and annotators, to the subtle anxiety produced by deepfakes and information chaos. Given the scale at which AI systems now mediate human emotion and attention, mental-health impacts need to be a core consideration, not a footnote under "social implications." Second, labor and ghost work. Somewhat tangentially, let me share this experience: I myself worked in the call center environment in the City of Medellín, Colombia, from January 2023 until July 2024, and my breaking point came when French-speaking customers from France and Canada displayed harassing behavior towards me due to my Latin American-inflicted accent when speaking French, and despite my graduate degree (MS in Translation - Spanish-English) from New York University and B2-level fluency in French, my employers at the time criticized my complimentary language efforts through the usage of Large Language Models and Machine Translation tools like DeepL and Linguee. Call me overconfident, but my professors did a phenomenal job at instilling a sense of ownership when it comes to the production and process of translated content through digital means. At any rate, The Dialogue should address the hidden workforce behind AI (data labelers, content moderators, crowdworkers) who are often in the Global South, poorly paid, and exposed to traumatic material with minimal protection.any serious governance framework must consider decent work standards, collective bargaining rights, and fair distribution of value along the AI supply chain. Third, media integrity and democracy in an age of synthetic media. While human rights and safety are mentioned, there is no explicit focus on how AI reshapes public discourse: targeted disinformation, automated propaganda, deepfake political content, and the erosion of trust in authentic journalism. For entities like ours, which operate as a critical media voice, this is existential. These issues cut across all thematic pillars: they affect how we define safety, what kinds of transparency matter, how we design capacity-building, and which open-source practices are actually emancipatory rather than extractive. Making them explicit would help the Global Dialogue connect high-level governance discussions to the daily realities of the people who live, work, and struggle under AI-mediated regimes.

They're hitting Brazil and East Africa in ways we see daily from the CovfefeTV desks: through WhatsApp, synthetic media, and invisible labor. On safety, human rights, and transparency, Brazil is a test case for platform power. WhatsApp's opening to rival AI chatbots (after CADE pushed back on Meta's attempted ban) creates an explosion of bots in the same channel people use for family and politics, with very uneven safeguards. This is a chance for competition and local innovation, but also a risk of unregulated "therapy" bots, predatory companions, and political manipulation at massive scale. Oversight, explainability, and clear red lines for high-risk use (e.g., AI chatbots simulating sexualized minors, which Brazil has already moved against) are lagging behind. In Kenya and the wider African region, the biggest gap is around labor and disinformation. Nairobi and similar hubs supply ghost workers who label data and moderate horrific content for a few dollars an hour, while the platforms capturing the value remain elsewhere. Trauma compensation, collective bargaining, and enforceable standards for decent work are still the exception, not the rule. At the same time, AI-driven disinformation and synthetic media are rapidly eroding trust in elections and journalism... from deepfake political content in Europe to AI-generated falsehoods about coups and protests in Brazil and African democracies. Regulatory responses tend to oscillate between under-enforement by platforms and over-reach by states (internet shutdowns, criminalization of speech), leaving independent media and fact-checkers to plug the gaps with limited resources. The opportunity is that these governance failures are now highly visible. Brazil's antitrust rulings on WhatsApp and public debates on AI in democracy, along with African content-moderator organizing and regional work on disinformation, give the Global Dialogue real cases to learn from... if it chooses to listen outward, not just to the usual capitals.

The AI Dialogue can turn today's fragmented landscape into a sequenced, UN-anchored conversation that actually travels between capitals, companies, and communities. First, it can act as the political landing zone for the Independent International Scientific Panel's evidence. That means structuring each annual meeting around the Panel's report and forcing a response: What recommendations will states act on, how will companies adapt their practices, and where do civil-society warnings align or clash with government priorities? Second, it can serve as a bridge between regimes: OECD/GPAI, the EU's AI Act, African Union and Latin American initiatives, national frameworks in countries like Brazil, and sectoral standards (e.g., in health, education, media). Rather than duplicating work, the Dialogue can map where governance approaches converge, where they conflict, and how to make them interoperable for cross-border systems. Third, it can institutionalize meaningful participation from the Majority World and frontline communities, not just in side events but in agenda-setting, drafting groups, and follow-up. That includes youth, workers in the AI supply chain, independent journalists, and mental-health advocates. Finally, it can provide a public narrative and accountability space: a regular moment when the UN can tell the world, in plain language, how AI is being governed, who is lagging, and where urgent course corrections are needed. For media actors like us, this makes AI governance legible enough to cover, critique, and translate back to our audiences.

The AI Dialogue should intentionally connect to: UNESCO's Recommendation on the Ethics of AI and its readiness assessments and capacity-building for judges and public servants. The OECD AI Principles and the Global Partnership on AI (GPAI), which already generate tools, pilots, and implementation guides grounded in those principles. Regional efforts (e.g., Council of Europe's AI convention, AU and EU-LAC digital cooperation) and sectoral initiatives on media, disinformation, and digital public goods. The added value of the AI Dialogue is not to become yet another principles factory. Instead, it can: 1. Align and sequence: Offer a shared roadmap that says "who does what when" across these initiatives, anchored in the Global Digital Compact and SDGs. 2. Fill representation gaps: Bring in states and stakeholders that are not part of GPAI or OECD but are deeply affected by AI deployment, especially in the Global South. 3. Surface conflicts and trade-offs: Provide a space where civil society, technical experts, and governments can openly discuss tensions (e.g., between open models and safety, or between content moderation and freedom of expression) rather than smoothing them over. 4. Track follow-through: Establish light-touch monitoring of how commitments made in other fora are implemented, feeding this back into the Dialogue's annual cycle. Done well, the Dialogue becomes the coordination spine for global AI governance... without replacing the specialized, technical, and regional work already underway.

Different stakeholders bring different forms of power to the Dialogue: states bring mandates, companies bring infrastructure, and communities bring lived evidence of harm and opportunity. The format should reflect that. Member States should use plenary segments to set political direction, but most of the substantive work should happen in co-chaired multistakeholder breakouts (one Member State + one non-state co-chair per theme), as already envisaged, with clear deliverables per session. Civil society, worker organizations, media, technical experts, and communities from the Majority World should be embedded in: A standing Stakeholder Reference Group that helps shape agendas and background papers between sessions. Thematic breakouts where they can present evidence, not just "make statements," and respond to state and industry interventions. The private sector should contribute transparency on models and supply chains, pay into a participation support fund for underrepresented stakeholders, and accept that the Dialogue is a governance space, not a product showcase. Academia and technical communities can help translate between political concerns and implementable standards or benchmarks. Structurally, I would recommend: One high-level plenary to frame priorities and hear from the Scientific Panel. Parallel thematic clusters with co-chairs and agreed questions, feeding into a Roadmap. A closing synthesis session where co-chairs report back and identify areas for negotiation in the next cycle.

Today's global AI governance debates still skew toward governments and institutions from the Global North, large companies, and elite universities. Missing or under-heard voices include: Workers in the AI supply chain: data labelers, content moderators, and gig workers in places like Kenya, the Philippines, Brazil, and India. Indigenous peoples and traditional communities, whose lands, knowledge, and data are increasingly targeted by AI-driven extraction. People with lived experience of mental-health crises, disability, and institutionalization, who are most affected by "AI for mental health" and predictive risk tools. Independent journalists, local media, and fact-checkers, who see AI-driven disinformation and harassment first-hand but rarely shape governance debates. Youth from the Majority World, whose futures will be structured by AI but who have limited access to decision-making spaces and technical education. Inclusion should not rely on one-off interventions. The Dialogue can: Reserve formal speaking slots and co-chair roles for representatives of these groups in each thematic cluster. Use the participation support fund to cover travel, visas, and connectivity for those without institutional backing. Recognize community-based training and certification programs as valid expertise, not just elite degrees. Partner with regional bodies and networks to run pre-Dialogues that feed directly into the agenda.

To avoid another static conference, the Dialogue can borrow from both multistakeholder internet governance and investigative journalism. Promising formats include: Case labs: small, moderated sessions where states, companies, and affected communities walk through a concrete AI incident (e.g., disinformation in an election, harm from a mental-health chatbot, mass layoffs tied to automation) and jointly map failures and fixes. Fishbowl dialogues: a rotating "inner circle" of speakers representing different stakeholder groups, with empty chairs reserved for underrepresented voices to join the inner circle during the session. Live governance simulations: role-play scenarios where participants must respond to an AI crisis (e.g., a deepfake of a head of state), forcing real-time negotiation between regulators, platforms, media, and civil society. Regional hubs connected virtually to Geneva/New York, allowing people in Nairobi, São Paulo, or Jakarta to feed live input into plenary discussions. Evidence galleries: curated exhibits (physical and digital) showing testimonies from workers, communities, and journalists affected by AI systems... something participants must walk through on their way into negotiations. These formats make it harder to ignore uncomfortable evidence and easier to see trade-offs. They also mirror how AI is actually experienced: as concrete incidents, not abstract principles.

Effective AI governance is already emerging in a few concrete policies and practices that the Dialogue can lift up and replicate. On the regulatory side, the EU AI Act is a landmark example: it classifies AI systems by risk and imposes strict obligations on "high-risk" uses, including mandatory risk assessments, high-quality datasets to reduce discriminatory outcomes, logging for traceability, detailed documentation, clear user information, human oversight, and robustness/cybersecurity requirements. It also introduces transparency duties for systems like chatbots so people know when they are talking to a machine, which is crucial for democratic debate and mental-health use cases. Brazil is moving in a similar risk-based direction with its AI Bill and sectoral rules in health. Even before a general AI law, Brazil's data-protection framework (LGPD) and ANPD hae been used as guardrails for health-related AI, while recent medical-council rules explicitly allow AI as decision support but require human supervision and the patient's right to refuse. This combination of baseline data protection plus sector-specific supervision and mandatory human oversight is a model other countries can study. Internationally, UNESCO's Recommendation on the Ethics of AI provides a human-rights-anchored framework with concrete policy action areas (data governance, environment, gender, education, health) and implementation tools such as readiness assessments and ethical impact assessments for high-risk systems. The Global Partnership on AI (GPAI) translates principles into pilots and policy tools across responsible AI, data governance, future of work, and innovation, offering a living lab for what works in practice. Finally, inside organizations, we see emerging good practice in embedding governance into the AI lifecycle: cross-functional oversight bodies, model documentation (model cards), mandatory bias and robustness testing before deployment, AI registries, and continuous monitoring using policy-as-code platforms. These show that AI governance can be operational, not just aspirational.