Green Transformation and Sustainability Network (GXS)

A successful first Global Dialogue on AI Governance would deliver three outcomes. First, it would move beyond restating principles to articulating a shared problem diagnosis and a small set of concrete, time‑bound priorities, anchored in human rights and the Sustainable Development Goals and informed by the Independent International Scientific Panel on AI's inaugural report. This includes recognising asymmetries in power, compute, data and voice between and within countries, and explicitly committing to narrow the global AI divide. Second, it would create practical instruments that states and stakeholders can immediately use: for example, a voluntary "menu" of tools such as AI and human‑rights impact assessment templates, model transparency and incident‑reporting practices, and options for public AI registers and regulatory sandboxes that different legal systems can adapt. For civil society and research actors, success means gaining predictable avenues to feed evidence, community experience, and independent evaluation into these instruments and future iterations of the Dialogue.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models
• Interoperability of governance approaches

For GXS AI Governance Lab, the four selected themes are tightly interconnected and central to our mission. 1. Social, economic, ethical, cultural, linguistic and technical implications of AI AI systems reshape economies, labour markets, information ecosystems and cultural expression, often in ways that amplify existing inequalities and marginalise certain communities and languages. Civil-society and research organisations are uniquely placed to surface these real-world effects, including on indigenous data sovereignty, gender and racial justice, and linguistic diversity, and to ensure that governance frameworks reflect local norms rather than importing a narrow set of values. 2. Transparency, accountability, and human oversight Without access to information about how AI systems are designed, trained, deployed and governed, neither regulators nor communities can meaningfully contest harms or demand redress. Effective oversight requires explainability appropriate to context, public registers, independent auditing, and clear allocation of responsibility across the AI lifecycle-not just "human in the loop" on paper. 3. Interoperability and compatibility of governance approaches Today's patchwork of national and sectoral frameworks risks fragmenting markets and leaving individuals unevenly protected. Interoperable approaches-grounded in international law and human rights yet flexible enough for diverse legal traditions-are essential for cross-border accountability, trade, and cooperation, particularly for smaller states that must navigate multiple regimes simultaneously. 4. Open-source software, open data and open AI models Open models and datasets can democratise access to AI capabilities, support local innovation and scrutiny, and reduce dependence on a small group of frontier firms. At the same time, they raise specific safety, security and misuse challenges that require tailored safeguards, licensing choices and shared evaluation. As a civil-society research lab, we see this as a critical area where multistakeholder guidance is urgently needed to balance openness, equity and risk.

Several cross-cutting and emerging issues deserve more explicit attention. First, AI and power asymmetries: Governance must grapple with structural imbalances in who designs, finances and controls advanced AI systems, and who bears the externalities. This includes concentration of compute and data resources, domination of a few languages and cultures, and the limited ability of many states and communities to negotiate terms with large providers.

Vietnam and Southeast Asia are experiencing rapid AI adoption, especially in digital government, finance, and platform economies, but governance is evolving more slowly and unevenly. Vietnam's new AI‑centred digital government blueprint and 2030 AI strategy aim to build a smart, AI‑driven state and position the country among the region's AI leaders, yet institutional capacity, data governance, and skills remain work in progress. On social, economic, ethical, cultural, linguistic and technical implications, AI is already shaping labour markets, public services and information ecosystems in ways that can deepen existing inequalities if not guided by robust safeguards and inclusive consultation—especially for rural communities, informal workers, women and ethnic minorities. At the same time, there is real opportunity to use AI to expand access to healthcare, education and climate‑resilient agriculture, provided community needs and local languages are embedded from the start. On transparency, accountability and human oversight, Vietnam and many ASEAN states are still developing systematic approaches to algorithmic impact assessment, public registers, independent auditing and redress. This creates risks of opaque deployment of AI in areas like social protection, credit scoring or public security, but also an opportunity to "build it right the first time" by drawing on global best practice and regional cooperation.

The AI Dialogue can become the backbone of a more coherent, equitable international cooperation architecture for AI governance in three ways. First, it can translate high‑level commitments from the Global Digital Compact and the "Governing AI for Humanity" report into a predictable, annual multistakeholder process with clear milestones. This includes aligning work of the Independent International Scientific Panel on AI, a future UN AI office, and existing regional and multilateral initiatives so that states, civil society, and researchers know how evidence and proposals will be considered over time. Second, the Dialogue can institutionalise cooperation on capacity‑building and knowledge sharing, especially for developing countries. This means going beyond ad hoc panels to broker concrete partnerships on regulatory training, shared evaluation infrastructure, and access to compute and open models in line with human‑rights and sustainability goals. Initiatives such as the Quantum Nexus Initiative and GXS AI Governance Lab can contribute practice‑oriented curricula, open simulation environments, and governance toolkits that support regulators, practitioners and communities in the Global South.

The AI Dialogue should consciously build on and connect to existing governance efforts—UNESCO's Recommendation on the Ethics of AI, the OECD AI Principles, GPAI, regional strategies (e.g., EU, AU, ASEAN), and UN system work under UNDP, ITU, UNESCO and the World Bank—rather than duplicate them. It should also draw on the High‑Level Advisory Body's proposals for a scientific panel, capacity‑development network, standards exchange and future UN AI office. The added value of the Dialogue lies in its ability to: Connect silos: provide a single interface where states, UN entities, standards bodies, development banks, and multi‑stakeholder partnerships can compare approaches, identify gaps and overlaps, and work toward interoperable, human‑rights‑centred governance.

Different stakeholders can contribute most effectively if the Dialogue is structured around clear roles and predictable entry points. - Member States should set priorities, share regulatory experiences, and commit to concrete follow‑up actions, including resourcing independent oversight and opening space for civil‑society participation at national level. -International organisations and development banks can map existing initiatives, align capacity‑building offers, and support a Global Capacity Development Network and shared evaluation infrastructure.

Several perspectives remain structurally underrepresented in global AI governance debates: - Communities in the Global South, especially least‑developed countries and small island developing states. - Grassroots organisations working on labour rights, informal work, disability, gender, indigenous and minority rights, and environmental justice.

To foster meaningful and dynamic engagement, the Dialogue could experiment with: - Fishbowl dialogues, where seats rotate between states, companies, civil society, and researchers, keeping discussions focused and interactive. - Case clinics built around concrete deployments (e.g., public‑sector AI project, open‑source model), with mixed teams diagnosing risks and proposing governance responses in real time. - Community evidence hearings, where affected groups and frontline practitioners present short testimonies followed by expert reflection on governance implications. - Design sprints or policy labs, run in parallel to plenaries, where participants co‑develop templates for AI impact assessments, transparency registers, or capacity‑building plans—drawing on open infrastructures like those developed by QNI and GXS AI Governance Lab. - A digital participation layer (live polls, moderated Q&A, regional hubs connected online) to ensure meaningful input from those unable to attend in person.

Several existing frameworks illustrate effective approaches to AI governance, while initiatives like Quantum Nexus Initiative (QNI) and GXS AI Governance Lab show how such principles can be localised and operationalised in the Global South. At the normative level, UNESCO's Recommendation on the Ethics of AI combines high-level values with concrete tools such as ethical impact assessments and readiness assessments that states can adapt to their context. Singapore's Model AI Governance Framework for Generative AI provides a practical, risk-based approach for developers and deployers, translating principles into operational guidance on accountability, transparency, safety and security. The World Bank's "Global Trends in AI Governance" report distils emerging practice into a toolbox-industry self-governance, soft law, sandboxes and hard law-that policymakers can tailor to local conditions.