Internet Society

A productive first dialogue will create three tangible outcomes. The first is a common starting point of an inventory of AI governance gaps in all world regions based upon national and/or regional input gathered prior to the meeting which accounts for both high-and-low capacity country discrepancies as remedies to begin addressing. Secondly, it creates governing principle(s) with respect to how different countries may choose to implement the same or similar governing principles; this will bridge some of the currently unaddressed gaps in implementing those recommended governance principles as articulated in UNESCO's 2021 Recommendation and OECD AI Principles. Thirdly, it establishes an ongoing formal process for stakeholders to continue exchanging information and ideas between meetings – providing structure to avoid creating "one-time" events such as was demonstrated at the Paris AI Action Summit held in February 2025, where momentum was lost due to lack of follow through mechanisms. A public repository of national AI governance strategies must be created during the dialogue (modeled after the WTO Trade Policy Review Mechanism) but designed to work quickly and accessibly. The repository will also serve as a record of adaptation made to each nation's AI strategy, including how they have addressed issues related to elderly individuals being defrauded via AI-based schemes; how adolescents are impacted by AI-enabled platforms; and how linguistic representation gaps impact the ability of individuals to understand their rights as well as access their rights via the use of AI. These are examples of global issues that require solutions developed within regional parameters. Success will mean establishing trusted channels of communication among government officials, civil society groups, researchers, and industry representatives that allow for the sharing of evidence and coordination of efforts when AI innovations outpace regulatory development.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Protection and promotion of human rights

Safe, secure, and trustworthy AI is foundational: all governance depends on predictable, auditable systems. The NIST AI Risk Management Framework and EU AI Act advance domestic models but lack an internationally accepted risk classification methodology for cross-border deployments, creating accountability vacuums where harms disappear between jurisdictions. Transparency, accountability, and human oversight are urgent as automated decisions shape credit, employment, health, and immigration outcomes globally. Algorithmic feeds on consumer platforms operate without meaningful transparency toward users. Converging neuroscience evidence, including the longitudinal Adolescent Brain Cognitive Development Study, shows these systems affect prefrontal cortical development, attention regulation, and social cognition in adolescents, positioning algorithmic transparency as a matter of public health. The Council of Europe's Framework Convention on AI offers a cross-system accountability model. AI capacity-building tackles structural inequality directly, targeting not only technical practitioners but regulators, judges, civil servants, and consumer protection agencies. Older adult populations require culturally adapted digital literacy programs specifically designed to address AI-powered impersonation fraud through voice synthesis and personalization. Protection and promotion of human rights grounds the entire governance agenda in existing international law. The UN High Commissioner's framework, combined with emerging evidence on AI's effects across the full human lifespan, makes rights impact assessments a necessary standard in AI procurement and deployment cycles.

Three issues cut across all listed themes and require dedicated attention: First, the spread of synthetic media and AI-disinformation has now surpassed the speed at which electoral laws can be developed and implemented. As demonstrated during the concurrent 2024 elections worldwide, governments need the help of international bodies such as International IDEA and the Electoral Integrity Project to create effective regulatory frameworks to manage AI-generated political content. Second, AI-based scams using voice-cloning and deep-fakes have evolved from traditional low-level scams by becoming targeted based on an individual's family structure and use of their emotions. Data collected from the AARP Fraud Network and the Australian Scam Prevention Framework indicates that there is an explosion of AI-based scamming occurring against people aged 60 and above. Similar data exists demonstrating this is a global problem with different forms within culturally-specific societies (the manner in which AI-created voice utilizes family relationship patterns differs significantly among kinship systems in Latin America, Sub-Sahara Africa, South East Asia and Southern Europe). Therefore, solutions will require collaboration among consumer protection organizations, telecommunications regulators and community-based organizations with local presence. Thirdly, the developing brains of young people have been impacted negatively by exposure to AI-generated platform optimization systems designed to maximize user-engagement. According to evidence compiled in the 2024 U.S. Surgeon General Report on Youth Mental Health and the Adolescent Brain Cognitive Development Study, while still forming, the cognitive structures of adolescents' brains are being affected by interactions with optimized-for-engagement algorithms. Australia established a minimum age law regarding digital products, and the UK established an "Age Appropriate Design Code" - comparative studies examining how cultural context and family structure influence these negative impacts would precede any suggestions or recommendations for a one-size-fits-all approach globally.

Latin American countries and the Caribbean region illustrate a "governance deficit" within the current AI moment as well as an opportunity for advancement. The countries of Brazil, Chile, Colombia and Argentina all have growing AI research communities and have developed national AI strategies; Chile's 2023 National AI Policy and proposed Brazilian law on AI can be seen as first steps toward development. However, there continues to exist a structural issue: large-scale deployments of AI are being created by people outside of Latin America; they are being trained and regulated in places other than Latin America, with values and assumptions about data usage that do not necessarily match those in Latin America. There exists the largest governance deficit when it comes to the use of public-sector AI. There are many examples throughout the region where AI is being used to automate social-benefit allocations, police operations and judicial assistance. Yet, few of these systems utilize procurement practices that require disclosure or provide assessment of the impacts associated with their algorithms. AI-enabled financial fraud targeting older adults is a rapidly growing challenge. Chile's SERNAC has begun incorporating digital fraud into consumer protection enforcement, but the regulatory framework has not caught up with the technical sophistication of AI-powered impersonation scams. The intergenerational trust dynamics characteristic of family communication in the region make voice-cloning attacks particularly effective and harmful in this specific cultural context. Finally, linguistic disparities exacerbate these issues. Compared to Indigenous languages, Spanish and Portuguese are far better represented within commercial AI applications. As a result, the reliability gaps that exist within AI-assisted public services directly impact millions of people throughout the region. Therefore, multilateral funding of representative data infrastructure and culturally-grounded methodology for evaluating AI effectiveness are critical components to addressing the gaps created by individual national governments.

The Dialogue may fill an important gap left by no existing forum as a vehicle for formalized and evidence-based dialogue among governance regimes being developed in parallel without adequate interaction. The EU AI Act, India's emerging framework, Brazil's AI Bill, the U.S. executive orders on AI, and the AU's AI continental strategy are each developing at differing paces based on differing underlying premises. None of the current forums (OECD, G7 Hiroshima process, Bletchley-Seoul-Paris Summit Series) have the mandate to provide a common deliberative space for all of these actors with equal status. This gives the Dialogue a competitive advantage due to its universal membership basis within the UN system. Therefore it has the ability to engage those states that are major consumers of AI (e.g., most of sub-Saharan Africa, Central Asia, and the Pacific Islands) who are not part of the OECD or G7, creating a legitimacy foundation for any guidance or standards that arise. Concretely, the Dialogue could establish a voluntary peer review function through which countries submit their AI governance approaches for structured feedback from other member states and expert bodies. This would generate learning without requiring treaty-level commitments and build the mutual trust needed for eventual convergence on shared standards. The ITU's role in spectrum governance offers a historical precedent for how a UN body facilitates technical coordination without displacing national sovereignty. The Dialogue should also develop explicit capacity to address governance challenges with asymmetric cultural expressions (including elder fraud and adolescent platform harms) that require globally coordinated technical standards combined with locally adapted implementation frameworks.

A number of projects have laid some strong groundwork for future efforts. The OECD's AI Policy Observatory has compiled what is arguably the best collection of comparative databases on all of the national AI policy regimes to date. Thus, it would make sense to formally integrate this information into the Dialogue's knowledge base. UNESCO's Recommendation on the Ethics of Artificial Intelligence was approved in 2021 by 193 countries. This recommendation establishes a universal ethical framework; and also sets up an implementation monitoring process which can serve as a model to track how effectively governments are implementing their governance commitments in the area of artificial intelligence. The Global Partnership on AI, now integrated into the OECD, developed working group consensus on responsible AI, data governance, and AI in the workplace. The Dialogue should commission similar groups on gaps not yet covered, including AI-enabled fraud against vulnerable populations and AI's role in adolescent digital environments. The Council of Europe's Framework Convention on AI is the first binding international treaty on the subject. Its drafting involved observer states from outside Europe, including the United States, Canada, and Japan, demonstrating that multilateral AI treaty-making is feasible. Tracking its implementation provides the Dialogue with a live experiment in binding governance. At the regional level, ECLAC has worked on developing principles for digital transformations in Latin America, while the African Union has developed its AI Continental Strategy which provides governance ideas/ideas/thinking that complements other thinking developed mainly in the Global North.

The structure of the Dialogue must mirror the comparative advantages of each of the participant groups. The governments provide legal authority and a history of implementing policy and should anchor the plenary sessions with formal position statements which will be available online and searchable. civil society organizations provide accountability and closeness to the impacted populations and should have guaranteed speaking slots during thematic sessions. The technical research communities provide evidence based on the capabilities and failure modes of systems as well as methodology for evaluating them and should facilitate structured expert panels with defined mandates to produce advisory outputs. Private sector contributions are most valuable when specific and verifiable. Companies should be invited to share documented governance practices (including incident reports, red-teaming results, and deployment impact assessments) in structured formats that allow comparison across organizations and over time. Consumer protection agencies, gerontological societies, pediatric and adolescent health research networks, and youth-led organizations represent stakeholder groups whose expertise is directly relevant to AI governance but who are rarely present in these forums. The Dialogue should actively solicit their structured contributions. The two-session structure creates an opportunity for deliberate sequencing. Geneva 2026 could focus on diagnosis: mapping governance gaps, hearing from diverse stakeholders about ground-level impacts, and identifying areas where international agreement is most feasible. New York 2027 could focus on action: presenting draft frameworks, reviewing pilot initiatives launched after Geneva, and making commitments measurable through agreed indicators. Between sessions, regional consultations convened by UN regional commissions would ensure that both plenary sessions draw on geographically distributed input rather than only from delegations that can sustain year-round Geneva presence.

Several communities are systematically excluded from global AI governance discussions. Indigenous peoples face AI systems that misclassify their languages, misrepresent their cultural heritage in training data, and inform resource extraction decisions affecting their territories, yet have almost no formal representation in governance forums. The UN Permanent Forum on Indigenous Issues could serve as a structural bridge, and the Dialogue should establish a dedicated consultation mechanism rather than treating Indigenous communities as a subset of general civil society. In terms of demographics, the fastest growing population to be impacted by AI-enabled negative consequences includes older adults. Older adults are being negatively impacted by sophisticated forms of financial fraud that utilize voice cloning technology and deepfake technology. HelpAge International and Age International are organizations that have first-hand knowledge about how these systems impact older adult populations in various cultural and economic contexts. Because regional vulnerability patterns vary significantly due to differences in: the structure of older adults' financial dependency; norms surrounding the way families communicate with one another; and access to digital infrastructure, it is essential that the lived experiences of older adults are considered primary sources of evidence. In addition, children and adolescents are also significantly impacted by AI systems and yet represent virtually no one in global governance forums. Youth-led organizations, networks of schools, and adolescent health researchers (including those focused on the intersection of developmental neuroscience and digital environments) should have formal avenues for input into the decision-making processes of the Dialogue. Furthermore, governance responses must be adapted to the local education system and family structures. Workers in the AI supply chain, particularly data annotators and content moderators in Kenya, Uganda, and the Philippines, experience governance failures as daily working conditions. The International Labour Organization provides an institutional entry point for their meaningful inclusion in the Dialogue.

The Dialogue needs to go beyond the typical model of plenaries and side events and instead create new forms of knowledge generation. Governance deliberation structures (structured deliberation) derived from the OECD's research on deliberative democracy can provide a vehicle to bring together random samples of citizens from various countries to formulate recommendations related to specific AI governance issues. There is already evidence to suggest that citizen assemblies in Ireland and France have been effective in dealing with complex technical and ethical issues. A transnational version of this format tailored specifically to address AI governance issues would be innovative and democratically justified. Governance red-teaming sessions could invite multidisciplinary teams to stress-test proposed frameworks by modeling how they would apply to specific AI deployment scenarios in healthcare, criminal justice, and border control. The outputs would be concrete and implementation-focused, surfacing challenges before frameworks are finalized. Regional asynchronous dialogues that take place between Geneva and New York sessions utilizing structured online platforms with facilitated translation capabilities allows those from regions/countries who do not have access to Geneva-based models to participate in substantive contributions. Dynamic coalitions from the Internet Governance Forum's format provides a useful precedent for continued thematic involvement at the intersession level. A public evidence repository where researchers, civil society groups, and affected individuals can submit documented cases of AI governance successes and failures would ground discussions in empirical reality spanning diverse cultural and economic contexts.

Effective AI governance is already being built in pieces across different legal systems, sectors, and scales of government. The task for the Dialogue is to connect these pieces into a coherent global architecture. For example, Singapore's Model AI Governance Framework illustrates how a principles-based approach becomes operational through sector-specific implementation guides covering finance, healthcare, and human resources. Its voluntary structure enabled rapid adoption while generating documented practice sufficient to inform future mandatory standards. Furthermore, the EU AI Act's risk based classification system creates a scalable architecture that others jurisdictions are already adapting. The compliance testing requirements and the EU AI Act's proposed AI Office to coordinate the use of AI will provide live experimentation examples of implementing AI governance at scale. Data from these implementations will be available prior to Geneva 2026. Regarding elder protection in Australia, their Scam Prevention Framework (which requires banks, telecommunications companies, and platforms to develop strategies to prevent, detect and stop the flow of funds for AI powered scams) and the AARP's work with GSMA and telecom companies on developing algorithms to identify AI-based fraud exemplify that the public-private partnerships between governments and industries can protect citizens at-scale. However, the common thread of success across both case studies indicates that successful responses to AI threats require culturally relevant solutions - family communication patterns, financial literacy baselines, and levels of institutional trust differ significantly among nations, therefore frameworks created in high-income nations require deliberate adaptation. Finally, Amsterdam and Helsinki's algorithmic registries have demonstrated that municipalities can implement transparent reporting requirements for local AI activity - these are technically possible and politically acceptable. Additionally, the open-source Algorithm Audit Platform used in the Netherlands, provides an audit process for independent evaluation of AI that can be used by civil society and government auditors without access to proprietary model weights.