IRIS Sustainable Development

A successful first Global Dialogue on AI Governance wouldn't just be about agreeing on principles, those are already widely shared. What would really matter is whether it changes how we think about who is affected by AI, and how. One important outcome would be the operationalization of intersectionality within AI governance frameworks. At present, AI-related risks are typically framed in broad and discrete categories, such as bias, safety, or transparency, without sufficient attention to how these risks intersect and compound across different social identities and structural conditions. In practice, however, harms are rarely experienced in isolation. AI systems that perform adequately at an aggregate level may still produce systematically unequal outcomes for specific groups, such as women in rural contexts, persons with disabilities, or socioeconomically disadvantaged populations. A meaningful dialogue would therefore promote governance approaches that explicitly account for these overlapping vulnerabilities, ensuring that assessment, design, and oversight mechanisms capture the complexity of real-world impacts rather than relying on generalized or average-case evaluations. This connects to a broader point also reflected in the Pact for the Future: technology can easily reinforce existing inequalities if governance doesn't actively counter them . So success would mean asking harder questions, not just "is AI beneficial?" but "beneficial for whom, and at whose expense?" Another indicator of success would be who is able to shape the conversation. If participation remains dominated by governments and large technology firms, the resulting outcomes are likely to overlook critical perspectives. Ensuring the inclusion of those who are typically positioned as subjects of these systems, young people, communities from the Global South, and marginalized groups, would significantly strengthen the process, particularly if they are actively involved in shaping solutions rather than merely consulted. Finally, it would matter whether the dialogue produces mechanisms for measuring unequal impacts, rather than focusing solely on aggregate progress. Without such approaches, it becomes far easier to overlook which groups are being left behind.

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

The selected priorities reflect a focus on equity, accountability, and the societal impacts of AI, particularly in relation to existing structural inequalities. First, AI capacity-building is essential to ensure that all countries and communities can meaningfully participate in the development and governance of AI systems. Without targeted investment in skills, infrastructure, and institutional capacity, especially in the Global South, there is a risk of deepening existing technological and economic divides. Second, the social, economic, ethical, cultural, linguistic and technical implications of AI are central because AI systems are not neutral. They are embedded in social contexts and can reinforce or reshape existing power dynamics. Particular attention is needed to how AI affects marginalized groups, including through language exclusion, cultural bias, or unequal access to opportunities. Third, transparency, accountability, and human oversight are critical for ensuring that AI systems remain subject to democratic control. This includes not only technical transparency, but also clear lines of responsibility, effective oversight mechanisms, and the ability to challenge or remedy harmful outcomes. Across these priorities, a common concern is the need to move beyond abstract principles toward practical implementation. This includes developing tools to assess and monitor unequal impacts, integrating intersectional perspectives into governance frameworks, and ensuring that affected communities are meaningfully involved in shaping AI policies. Together, these areas support a more inclusive and responsible approach to AI governance that prioritizes fairness, participation, and long-term societal resilience.

First, intersectionality and cumulative harms remain underdeveloped in current frameworks. Existing approaches tend to assess risks in isolation (e.g., bias or safety), but in practice, harms often overlap across axes such as gender, race, disability, and socioeconomic status. Without explicitly integrating intersectional analysis, governance efforts risk overlooking those most affected. Second, power asymmetries in AI development and governance are a critical but often implicit issue. A small number of actors, primarily large technology companies and a few states, shape the direction of AI systems globally. This raises concerns about representation, agenda-setting, and whose values are embedded in AI. Addressing this requires not only participation, but meaningful redistribution of influence in decision-making processes. Third, data governance and data justice deserve greater emphasis. AI systems depend on vast datasets, yet questions around data ownership, consent, extraction (especially from the Global South), and benefit-sharing remain insufficiently addressed. This includes the risk of reinforcing extractive digital economies. Fourth, the environmental and resource impact of AI is an emerging concern. The energy consumption of large-scale models, as well as the material footprint of digital infrastructure, raises sustainability issues that intersect with global inequalities. Finally, measurement and evaluation of impacts remains a gap. While many frameworks set out principles, there is limited agreement on how to assess real-world outcomes, particularly in terms of unequal impacts across different groups. Addressing these issues would strengthen AI governance by making it more grounded in real-world power dynamics, distributive effects, and long-term sustainability.

In terms of social and cultural implications, there are concerns that AI systems used in areas such as welfare, migration, or education may not adequately reflect the realities of marginalized groups. Even in a high-income context like Sweden, issues related to bias, language, and representation persist, particularly affecting migrants, minorities, and persons with disabilities. NGOs often encounter these impacts directly in their work but lack formal channels to influence system design. Another challenge is the limited transparency and accountability of AI systems, especially when procured from private vendors. NGOs working on rights-based issues report difficulties in accessing information about how automated decisions are made, which complicates advocacy and the ability to support affected individuals.

The AI Dialogue can play a substantive role in advancing international cooperation by bridging normative commitments and coordinated implementation across jurisdictions. First, it can facilitate structured convergence among diverse governance approaches. Rather than pursuing full harmonization, the Dialogue can help identify shared principles, risk classifications, and procedural standards that enable interoperability. This is particularly important given the current fragmentation of regulatory frameworks, which risks creating compliance challenges, uneven protections, and regulatory arbitrage. Second, the Dialogue can strengthen inclusive and multi-level governance. AI governance increasingly involves a wide range of actors beyond states, including civil society, the private sector, and international organizations. By systematically incorporating perspectives from underrepresented regions and groups, particularly from the Global South, the Dialogue can help address asymmetries in agenda-setting and ensure that governance reflects diverse social and economic realities.

The AI Dialogue arrives at a critical juncture in 2026, as governance shifts from high-level principles to implementation and enforcement. To maximize impact, it should build on and connect existing initiatives while addressing gaps between them. Among key foundations, the UN Independent International Scientific Panel on AI provides an essential evidence base on global risks and impacts. The Dialogue can translate this scientific input into policy-relevant priorities. The G7 Hiroshima AI Process (HAIP), with its reporting framework, offers practical tools for transparency and safety disclosures that could be expanded beyond G7 members. The OECD AI Policy Observatory functions as a central data hub on AI incidents and national strategies, supporting monitoring and comparative analysis. At the regulatory level, the EU AI Act, becoming fully applicable in August 2026, sets an influential benchmark for risk-based governance that other jurisdictions are already engaging with. However, these initiatives remain partially fragmented by geography, mandate, or level of development. The added value of the AI Dialogue lies in its ability to connect these layers into a more coherent global governance ecosystem.

Different stakeholders can contribute by leveraging their distinct roles and expertise. Governments can share regulatory approaches and implementation challenges, while the private sector can provide technical insights and operational practices. Civil society and NGOs are essential for bringing evidence of real-world impacts, particularly on human rights and inequality, and academia can contribute research and evaluation tools. International organizations can support coordination and continuity across processes.To enable this, the Dialogue should remain genuinely multi-stakeholder and interactive. The proposed structure, combining plenary sessions with thematic breakouts, is a strong foundation, but its effectiveness depends on how discussions are conducted.

Communities from the Global South are often included in discussions, but rarely have equal influence over agendas or decisions. As a result, governance frameworks may fail to reflect diverse economic conditions, languages, and infrastructural realities. Similarly, civil society organizations and NGOs—especially smaller or grassroots groups—face constraints due to limited funding and technical expertise, despite being closely engaged with the everyday impacts of AI in areas such as welfare, migration, and labour. Marginalized and intersectional groups, including women, persons with disabilities, racial and ethnic minorities, and low-income communities, are also insufficiently represented. Their experiences are often mediated through others, which can dilute or overlook more complex, overlapping forms of harm. Although youth and future generations are increasingly recognized in these discussions, they still lack stable and institutionalized roles, despite having a long-term stake in how AI systems are governed. Addressing these gaps requires moving beyond consultation toward genuine participation and co-creation. This involves providing financial and technical support, ensuring inclusion in core decision-making spaces, and establishing structured mechanisms, such as stakeholder panels or co-design processes, that enable meaningful influence.

One approach could be deliberative mini-publics or citizen assemblies integrated into the Dialogue. A diverse group of participants, selected to reflect different regions and social backgrounds could engage in structured deliberation and present recommendations. This would introduce perspectives beyond institutional actors and strengthen legitimacy. Reverse panels could also be an effective approach. More precisely, instead of policymakers questioning stakeholders, communities and NGOs question governments and companies directly shifting power dynamics and creates accountability in real time.

Several existing policies and practices offer concrete approaches to effective AI governance by combining regulation, accountability, and practical tools. The EU AI Act is a key example of a risk-based framework, classifying AI systems by potential harm and applying stricter requirements to high-risk uses. It moves beyond principles toward enforceable obligations. At the implementation level, algorithmic impact assessments (AIAs), such as those used in Canada and FRIAs help identify and mitigate risks before deployment, making governance more proactive. For accountability, independent audits and certification schemes are emerging to assess compliance with fairness and transparency standards. Similarly, model documentation practices (e.g., model cards, datasheets) improve transparency by providing information on how systems are developed and used. Multi-stakeholder platforms also contribute. The OECD AI Policy Observatory supports data-sharing and monitoring, while the G7 Hiroshima AI Process promotes common reporting practices. Finally, participatory approaches, including co-design and public consultations, help ensure that governance reflects real-world impacts.