Institute for Governance, Policies and Politics

The first Global Dialogue would be a success if it delivers concrete, forward-looking outputs rather than aspirational statements. Specifically: 1. A Co-Chairs' summary that translates the Panel's report into 3–5 prioritized, actionable recommendations with clear timelines and responsible actors. 2. Launch of at least two new multi-stakeholder initiatives: one focused on practical interoperability testing frameworks for governance approaches and another on scaling open-source AI models and compute access for developing countries. 3. Explicit commitments from Member States and major private-sector players to concrete capacity-building measures that begin closing digital divides within 12–18 months. 4. Agreement on a lightweight, iterative follow-up mechanism (annual or biennial Dialogue with rotating thematic focus) to keep governance adaptive rather than static. 5. Demonstrable evidence that the Dialogue accelerated practical cooperation for example, new open benchmarks for safety and trustworthiness that are adopted across regions. Success would be measured not by consensus language but by whether the Dialogue leaves the world with more open tools, clearer shared standards and tangible bridges across divides, enabling AI to advance the SDGs faster while remaining safe, transparent and broadly accessible. This would establish the platform as a genuine accelerator of responsible innovation.

• Safe, secure and trustworthy AI
• AI capacity-building
• Interoperability of governance approaches
• Open-source software, open data and open AI models

1. Safe, secure and trustworthy AI are the core technical and ethical prerequisites. Without reliable safety mechanisms, all other efforts risk amplifying harm or eroding public trust. 2. AI Capacity-building: Capacity gaps and bridging divides address the stark inequality in access to compute, data, skills and infrastructure. This is the most urgent development divide today and leaving it unaddressed will concentrate AI power and benefits in a few countries, undermining the Dialogue's SDG and inclusivity goals. 3. Interoperability of governance approaches is essential because fragmented, incompatible national or regional rules create friction, raise compliance costs and discourage cross-border collaboration. Compatible frameworks allow innovation to flow while maintaining baseline protections. 4. Open-source, open data and open models act as a powerful multiplier for the other three: openness enables broader scrutiny and collective safety improvements democratizes access and builds capacity faster and makes true interoperability feasible by allowing shared standards and transparent implementations. These four priorities represent the foundational enablers and primary bottlenecks of responsible AI. Deeply interconnected, they exert a more significant influence on real-world outcomes over the short-to-medium term than any other factors. By focusing here, we build the technical, physical and cooperative infrastructure necessary to support broader societal benefits and human rights. This approach prioritizes actionable leverage points designed to yield measurable progress by the next Dialogue, moving beyond purely aspirational discourse.

Yes, two critical cross-cutting issues are not explicitly captured: 1. Environmental sustainability and resource intensity of AI systems. The energy, water and material demands of frontier training and inference are growing exponentially and already rival those of entire countries. While technical implications are mentioned, there is no dedicated focus on green-AI metrics, sustainable compute infrastructure or reconciling AI's climate-solution potential with its own footprint. This issue intersects every listed theme: unsafe or unaccountable systems waste resources; capacity gaps widen if only wealthy actors can afford efficient hardware; open models can help diffuse efficient designs but only if sustainability is prioritized. 2. The need for agile, adaptive governance mechanisms. AI capability advances faster than traditional policy cycles. Static rules risk becoming obsolete or overly restrictive within months. The Dialogue should emphasize iterative processes, sunset clauses, continuous scientific review, real-time monitoring protocols and 'governance sandboxes', that keep oversight proportionate and up-to-date. This is cross-cutting: it affects interoperability (standards must evolve together), human rights protection (rights frameworks must keep pace), transparency/accountability (oversight tools must match new architectures) and open-source development (openness only works if governance can adapt without constant renegotiation). Addressing these would prevent the dialogue from solving yesterday's problems while tomorrow's challenges including resource constraints and regulatory lag to undermine its goals.

From the perspective of the global AI innovation sector, with a focus on responsible development and deployment, the four selected themes safe, secure and trustworthy AI systems, capacity building, interoperability of governance approaches and open-source software, open data and open AI models are shaping outcomes significantly. Significant Challenges 1. Governance gaps in these areas are enabling widespread misuse of open-source generative models. A striking example is the explosion of AI-based nudification applications. These tools, frequently built by fine-tuning publicly available open-source diffusion models, generate non-consensual intimate imagery with alarming ease. 2. The vast majority of deepfakes today are non-consensual pornographic content, disproportionately harming women and girls through image-based sexual abuse, sextortion and psychological trauma. The open nature of these models makes safety mechanisms easy to strip away, turning powerful tools intended for positive innovation into instruments of harm at global scale. 3. This is compounded by fragmented governance, which creates uneven standards and enforcement. Capacity gaps further exacerbate the problem as many regions lack both the technical expertise to implement safeguards and the regulatory frameworks to address downstream misuse, allowing harmful applications to proliferate unchecked in lower-resource environments. Significant Opportunities 1. Advances in open-source AI simultaneously offer enormous potential. Open models enable transparent safety research, rapid community-driven improvements and broader access to powerful tools essential for bridging capacity divides and building trustworthy systems. If the Global Dialogue promotes robust, harder-to-remove alignment techniques, standardized safety protocols, community norms against high-risk misuse (such as non-consensual imagery) and practical detection/watermarking tools, open-source AI can become a genuine global public good. 2. Targeted international cooperation on interoperability could establish baseline protections that travel with models across borders, while capacity-building efforts equip developing countries with responsible deployment capabilities rather than unchecked risks.

The Global Dialogue on AI Governance can serve as a unique, high-level platform to move beyond fragmented national approaches toward meaningful international cooperation. Its primary role should be to act as a bridge-builder that translates diverse perspectives into practical, interoperable outcomes. Specifically, the Dialogue can: Facilitate convergence on baseline standards: By bringing together Member States, the private sector, civil society, academia and technical communities, it can identify common minimum guardrails particularly for safe, secure and trustworthy AI systems, protection of human rights and prevention of high-risk misuse. A clear example is the urgent need to address the proliferation of open-source AI tools used for non-consensual intimate imagery. These tools, often created by stripping safety mechanisms from publicly available models, cause widespread harm, especially to women and girls. Promote interoperability of governance frameworks: Rather than seeking a single global treaty, the Dialogue can encourage compatible approaches that reduce regulatory fragmentation, lower compliance burdens and enable responsible cross-border collaboration and deployment. Promoting global cooperation on responsible innovation & techno-legal framework for interoperability. Support capacity-building and equitable access: It can mobilize concrete partnerships for sharing compute resources, open datasets and responsible open-source models, ensuring developing countries gain the skills and infrastructure needed to participate meaningfully in AI development not just as users, but as co-creators. Establish a durable follow-up mechanism: The Dialogue should institutionalize regular, evidence-based reviews informed by the Independent International Scientific Panel on AI, allowing governance to evolve with technological advances. By focusing on actionable outputs, joint initiatives, shared toolkits and measurable commitments rather than purely declaratory language, the Global Dialogue can become a catalyst for cooperative, adaptive and inclusive AI governance that advances the Sustainable Development Goals while safeguarding human dignity.

The Global Dialogue should actively build upon and connect with several established initiatives to avoid duplication and maximize impact: UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) and its Global AI Ethics and Governance Observatory the only global normative ethical framework adopted by 193 countries, offering readiness assessment tools and best practices on human rights, transparency and bias. AI for Good Global Summit (led by ITU with over 50 UN agencies) already providing action-oriented platforms for AI applications in sustainable development. OECD AI Principles and the Global Partnership on AI (GPAI) have strong foundations for trustworthy AI, risk management and multi-stakeholder collaboration. Hiroshima AI Process (G7) and the Bletchley–Seoul–Paris AI Safety Summit series which advanced international codes of conduct, safety testing and scientific reporting on advanced AI risks. Global Digital Compact and the Independent International Scientific Panel on AI ensuring scientific evidence directly informs policy discussions. Regional efforts such as the EU AI Act, ASEAN Responsible AI Roadmap, African Union Continental AI Strategy and capacity-building initiatives from China's Global AI Governance Action Plan and BRICS AI cooperation. As the first truly universal, inclusive UN platform with high-level governmental participation, it can bridge fragmented approaches by fostering interoperability of governance frameworks, promoting responsible open-source AI development while addressing misuse risks and scaling capacity-building for developing countries. It should deliver practical outcomes including shared safety protocols resistant to easy removal, detection/watermarking standards, community norms against harmful applications and concrete partnerships for equitable access to compute and skills. By linking scientific evidence from the Panel with multi-stakeholder action, the Dialogue can shift from voluntary principles toward adaptive, cooperative governance that advances the SDGs while protecting human dignity.

Stakeholder Contributions: The Global Dialogue's success depends on inclusive, meaningful participation from all actors: Member States: Provide policy priorities, national experiences and commitments to capacity-building and interoperability. They lead the high-level governmental segment and shape the Co-Chairs' summary. Private sector and technical community: Share best practices on safe and trustworthy AI systems, open-source model development and real-world misuse cases (such as AI-based nudification applications). They can demonstrate technical solutions like tamper-resistant safeguards, watermarking and responsible deployment toolkits. Civil society and academia: Highlight human rights implications, ethical concerns and impacts on vulnerable groups. They bring evidence on societal harms and ensure voices from developing countries and underrepresented communities are heard. International organizations (ITU, UNESCO, etc.): Facilitate knowledge-sharing, link to existing initiatives (e.g., UNESCO AI Ethics Recommendation, GPAI, AI for Good) and support the Scientific Panel's report integration. Recommendations for Format and Structure: To ensure dynamic, actionable outcomes, the two-day Dialogue should combine: 1. A high-level governmental segment for political commitment and Co-Chairs' summary. 2. Multi-stakeholder plenary with structured interventions and responses from the Scientific Panel. 3. Thematic break-out sessions clustered around the mandated topics (e.g., safe systems + open-source; capacity gaps + interoperability), using interactive formats such as expert panels, solution-focused workshops and matchmaking for new partnerships. 4. Dedicated side events or open mic slots for emerging issues, including practical demonstrations of misuse prevention tools and open AI safety practices. 5. Clear integration of the Independent International Scientific Panel on AI report as a foundational evidence base for all discussions. The structure should prioritize concrete deliverables such as voluntary pledges, toolkits, or follow-up working groups over general statements. Hybrid participation and regional preparatory inputs will enhance inclusivity and bridge digital divides. This format would transform the Dialogue into a catalyst for practical international cooperation.

Several critical voices remain underrepresented in global AI governance discussions, despite the Dialogue's commitment to inclusivity: 1. Global South governments, researchers, and communities (especially from Africa, Latin America, and parts of Asia): They often lack meaningful influence, serving more as data sources or end-users than co-shapers of norms. Over 100 countries have little to no voice in major initiatives. 2. Women and girls, particularly in the Global South, who face intersecting barriers in AI education, workforce participation, and exposure to specific harms such as gender-biased systems or non-consensual nudification and deepfake abuse. 3. Children and youth: Children are significantly underrepresented despite being among the most vulnerable to AI-generated harms, including non-consensual intimate imagery, manipulative content, and long-term impacts on their development, privacy, and mental health. Their perspectives are rarely considered in governance discussions. 4. Indigenous peoples and linguistically diverse communities: Their knowledge systems, cultural values, and low-resource languages are seldom reflected in training data or policy priorities. 5. Civil society organizations from marginalized or low-resource settings, persons with disabilities, and frontline communities affected by AI's social, economic, and environmental impacts. To include them effectively The Global Dialogue on AI Governance can lead by adopting a bottom-up approach that prioritizes regional inputs and reimagines value-based principles grounded in diverse cultural, social, and ethical contexts rather than top-down, one-size-fits-all frameworks. Concrete measures include: 1. Lowering barriers: Provide travel grants, translation services (including for low-resource languages), hybrid formats. Building on existing capacity-building by UNESCO, ITU and regional networks. 2. Targeted outreach and regional hubs: Organize preparatory consultations in underrepresented regions and partner with local networks like Masakhane, Deep Learning Indaba or women-focused AI initiatives. 3. Dedicated slots and matchmaking: Reserve speaking time and breakout sessions for Global South and civil society voices while facilitate 'matchmaking' between them and technical experts or funders. 4. Practical support mechanisms: Fund community-led research and case studies on local AI realities, integrate inputs from Indigenous and women's groups into the Scientific Panel's work and thematic discussions. 5. Ongoing accountability: Track representation metrics in reports and establish a lightweight 'Inclusion Working Group' for follow-up.

To move beyond traditional panel discussions and plenary statements, the Global Dialogue on AI Governance should adopt interactive, solution-oriented formats that encourage genuine exchange, cross-stakeholder collaboration and actionable outcomes especially on safe, secure and trustworthy AI systems, capacity gaps, interoperability and responsible open-source development. Recommended innovative formats include: 1. Solution Labs / Collaborative Workshops: Small, facilitated breakout groups mixing Member States, private sector (including open-source developers), civil society and technical experts. Participants co-design practical tools for example, a 'Responsible Open-Source AI Safety Playbook' addressing misuse risks like AI-based nudification applications through tamper-resistant safeguards and community norms. 2. 'Misuse-to-Mitigation' Case Clinics: Real-time, moderated sessions where stakeholders present verified misuse examples followed by rapid brainstorming of technical, policy and capacity-building responses. The Independent Scientific Panel provides evidence-based input. 3. Interoperability Matchmaking Sessions: Structured networking where governments and organizations pitch governance approaches and seek compatible partners for pilot projects on cross-border standards or shared compute/capacity initiatives. 4. World Café or Fishbowl Discussions: Rotating small-group conversations on thematic clusters, with rotating 'fishbowl' observers (including underrepresented Global South and women's voices) who later synthesize insights for plenary feedback. 5. Demo & Pitch Arena: Short, live demonstrations of open-source safety tools, detection/watermarking technologies or capacity-building platforms, followed by audience voting and partnership commitments. 6.Hybrid 'Rapid Response' Roundtables: Blending in-person and virtual participants for agile discussions on emerging issues, with real-time input from the Scientific Panel. These formats should be time-bound, moderated for balance and explicitly linked to deliverables. Combined with hybrid access, translation support and targeted invitations for underrepresented communities, they would transform the Dialogue into a dynamic platform for trust-building and practical cooperation rather than formal speeches.

Several promising policies, practices, and platforms already offer concrete solutions to AI governance challenges: 1.Open-Source AI Safety Practices: Initiatives such as Hugging Face's Safety Cards, Model Cards, and the BigCode project demonstrate how transparency, documentation, and community-driven evaluations can improve trustworthiness of open models. 2. Singapore's Model AI Governance Framework: A practical, sector-agnostic approach emphasizing human oversight, transparency, and accountability. It uses risk-based assessments and is widely adopted by companies in Asia. 3. EU AI Act: The world's first comprehensive horizontal AI regulation, which classifies systems by risk levels and imposes strict requirements on high-risk applications, including transparency and human oversight. 4. UNESCO Recommendation on the Ethics of AI (2021): The first global normative instrument providing ethical guidance on human rights, fairness, and sustainability, with implementation tools and capacity-building programs for developing countries. 5. Partnership on AI (PAI) and the Global Partnership on AI (GPAI/OECD): Multi-stakeholder platforms that develop practical toolkits for responsible AI deployment and risk management. The Global Dialogue should build upon these examples by promoting interoperability between different frameworks, encouraging the adoption of responsible open-source practices, and creating mechanisms to prevent misuse while preserving innovation. Special attention should be given to adapting these approaches to the needs of the Global South through bottom-up regional inputs and value-based principles that respect cultural diversity.