University of Birmingham, UK

The first Global Dialogue on AI Governance will be successful if it moves beyond broad principles and delivers a small number of concrete, monitorable outcomes. In my view, success would mean, first, agreement on a minimum global floor for high-risk AI: baseline commitments on safety testing, transparency, accountability and meaningful human oversight. This matters because AI diffusion is accelerating far faster than governance; UNCTAD estimates AI could generate $4.8 trillion in economic value, yet the benefits and risks are being distributed highly unevenly. Second, the Dialogue should produce a credible package on inclusion and capacity-building. At present, 118 countries are absent from major AI governance discussions, while Africa holds less than 1% of global data-centre capacity despite 18% of the world's population. A successful outcome would therefore include practical commitments on shared compute access, regulatory training, and technical support for developing countries. Third, success requires institutional continuity: a standing multi-stakeholder mechanism, inter-sessional working groups, and a public implementation scorecard so that commitments made in Geneva can be tracked before the 2027 session in New York. Finally, the Dialogue should establish at least one durable global public-good mechanism, such as an AI incident registry or shared safety-testing infrastructure, so that states can learn from failures rather than repeat them in isolation. In short, success is not a well-worded declaration. It is an actionable framework that is inclusive, rights-respecting, and capable of narrowing rather than widening the next great global divide.

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

Safe, secure and trustworthy AI is essential because deployment is outpacing oversight, especially in high-risk sectors such as health, education, employment, border control and public administration. Without a common safety floor, harmful systems can simply migrate to weaker jurisdictions. AI capacity-building is equally urgent because the global AI landscape is deeply unequal. 118 countries remain outside major governance discussions, and infrastructure is highly concentrated. If developing countries lack compute, skills and regulatory capacity, they will not shape AI governance; they will merely absorb its consequences. Protection and promotion of human rights must remain central because AI is already affecting privacy, equality, due process, freedom of expression and access to remedy. These are not hypothetical risks. They are current governance failures that require stronger safeguards grounded in international human rights law. Transparency, accountability, and human oversight is the operational bridge between principles and practice. People must know when AI is being used in consequential decisions, regulators must be able to scrutinise high-risk systems, and affected individuals must be able to seek meaningful human review. Taken together, these four priorities create a coherent agenda: build safe systems, ensure all countries can participate, protect fundamental rights, and make accountability real. That combination offers the strongest basis for legitimate and globally inclusive AI governance.

Yes. Several cross-cutting issues deserve explicit attention because they shape the success or failure of all the listed themes. First, AI's environmental footprint should be recognised more clearly. Large-scale AI systems are driving growing energy and freshwater demand through data-centre expansion. Governance that ignores environmental cost risks solving one problem while worsening another, especially for climate-vulnerable countries. Second, the Dialogue should explicitly address the compute and infrastructure divide. Capacity-building is important, but the deeper structural issue is unequal access to compute, cloud infrastructure, high-quality datasets and evaluation tools. Without addressing these foundations, many countries will remain dependent rule-takers rather than rule-shapers. Third, linguistic and cultural representation requires stronger visibility. AI systems continue to privilege high-resource languages and dominant cultural contexts, which can marginalise minority languages, indigenous knowledge and locally relevant social norms. This is both a development issue and a rights issue. Fourth, there is a need for global mechanisms for incident reporting and shared learning. At present, there is no universally recognised international registry for AI failures, harms or near-misses. Creating one would strengthen transparency, evidence-based policymaking and early warning capacity. Finally, the Dialogue should pay greater attention to institutional inclusion and implementation continuity. Multi-stakeholder participation, representation from the Global South, and follow-through between annual sessions are not procedural extras; they are conditions for legitimacy and effectiveness.

From the perspective of the UK higher-education and public-interest research sector, the governance gaps in AI are already having visible effects. The biggest challenge is regulatory fragmentation. The UK is still relying largely on non-statutory principles and targeted measures, while the EU AI Act has already entered phased application, with prohibited practices and AI-literacy duties applying from 2 February 2025 and GPAI obligations from 2 August 2025. For universities, researchers and cross-border partners, this creates compliance uncertainty, duplicated assurance work, and uneven expectations on safety, transparency and accountability. A second challenge is capability inequality. In our sector, institutions are expected to innovate quickly, but staff and students do not have equal access to tools, training or clear guidance. Jisc's 2025 research shows students are already using AI daily, yet their biggest concern is employability, alongside worries about misinformation, privacy, bias and unequal access. Russell Group universities have similarly stressed the need for AI literacy, academic integrity, and fairness where advanced tools sit behind paywalls. At the same time, the opportunities are substantial. The UK AI sector is growing rapidly: dedicated AI company revenues rose to £4.9 billion in 2024, AI-related employment reached 86,139, and inward investment projects were linked to more than £15 billion in capital investment. For universities and research institutions, this creates major opportunities in skills development, interdisciplinary research, public-service innovation and responsible commercialisation. The most significant need, therefore, is not less innovation but better-governed innovation: clearer international baselines, stronger human-rights safeguards, and more serious investment in capacity-building so that institutions can adopt AI confidently, ethically and inclusively.

The AI Dialogue can play a uniquely important role by serving as the first universal platform where all states, not only the most technologically advanced, can shape the rules, norms and institutional architecture of AI governance. Its comparative advantage is not to replace national or regional regulation, but to connect fragmented approaches through interoperability, shared standards and inclusive cooperation. First, it can help establish a minimum global baseline for high-risk AI: common expectations on safety testing, transparency, accountability and meaningful human oversight. This would reduce regulatory arbitrage, improve trust, and make it easier for countries with different legal systems to recognise each other's governance arrangements without forcing full harmonisation. Second, the Dialogue can strengthen international solidarity and capacity-building. Many countries still lack compute infrastructure, regulatory expertise and technical capacity to participate effectively in AI governance. The Dialogue can mobilise cooperation on shared compute access, regulatory training, open-source tools, and financing mechanisms so that developing countries become rule-shapers, not rule-takers. Third, it can create practical global public goods: for example, an international AI incident registry, shared safety-testing infrastructure, and common reporting templates for high-risk failures and near-misses. These would make cooperation evidence-based rather than purely declaratory. Finally, the Dialogue can give international cooperation continuity and legitimacy by establishing inter-sessional working groups, stronger interaction between the Scientific Panel and policymakers, and a multi-stakeholder mechanism with meaningful representation from civil society, academia, technical experts and affected communities, especially from the Global South. In that sense, the Dialogue can become not just a forum for discussion, but a durable mechanism for coordination, learning and implementation.

The AI Dialogue should build on, and connect, several initiatives that already provide important pieces of the governance architecture. OECD/GPAI offers a strong policy and expert community, and since its 2024 integration it has brought OECD and GPAI countries together "on an equal footing" to reduce duplication and advance safe, secure and trustworthy AI. UNESCO's Recommendation on the Ethics of AI and its Readiness Assessment Methodology (RAM) provide globally applicable ethical principles and practical tools for identifying national governance gaps. The Council of Europe AI Convention provides a legal framework anchored in human rights, democracy and the rule of law. The Hiroshima AI Process and its Friends Group have expanded the conversation on international guidelines and codes of conduct beyond the G7, including across the Global South. The ITU AI for Good ecosystem contributes standards, technical exchange and capacity-building, while the African Union's Continental AI Strategy shows the importance of regionally grounded, development-focused governance. The Dialogue should also work closely with the Independent International Scientific Panel on AI so that political negotiations are informed by credible scientific evidence. Its added value is that it can do what none of these mechanisms can do alone: provide a universal UN platform where all states can deliberate on AI governance, not just regional blocs or like-minded coalitions. The Dialogue can translate existing principles into a minimum interoperable global baseline on safety, transparency, accountability and human oversight; connect ethical frameworks with technical standards and capacity-building tools; and give underrepresented countries a real voice in shaping norms. It can also add continuity through inter-sessional working groups, implementation tracking and shared global mechanisms, such as incident reporting and safety-testing cooperation. In that sense, the Dialogue's core value is not to duplicate existing initiatives, but to align, legitimise and operationalise them at global scale.

Different stakeholders should contribute in ways that reflect their distinct responsibilities and comparative advantages. Member States should negotiate political commitments, report on implementation, and identify areas where national approaches can become interoperable. International organisations and regulators should bring comparative experience, technical assistance, and implementation support. The Independent International Scientific Panel should provide evidence syntheses, risk assessments, and options for action before each session. Private sector actors should contribute technical knowledge, safety-testing experience, and transparent reporting on risks, incidents and mitigation measures. Civil society, academia, and affected communities should play a substantive role in scrutinising impacts on rights, inequality, labour, language, culture and inclusion, especially from the Global South and underrepresented communities. In terms of format, the Dialogue should be designed as a genuine multistakeholder process, not a traditional sequence of formal statements. I would recommend: a high-level plenary for political direction; smaller thematic roundtables focused on negotiated outputs; and dedicated science-policy sessions where Heads of Delegation can engage directly with the Scientific Panel. Written submissions should remain open, accessible in all six UN languages, and linked to public summary papers so that stakeholder contributions visibly inform deliberations. To make participation meaningful, the Dialogue should establish a permanent multi-stakeholder advisory mechanism with balanced representation from civil society, academia, the technical community and affected groups. Between annual sessions, inter-sessional working groups should meet regularly to draft commitments, narrow differences, and prepare implementation options. Each Dialogue should then conclude with a public record of commitments and a scorecard reviewing progress made since the previous session. That structure would make the Dialogue more inclusive, evidence-based and action-oriented.

The most underrepresented voices in global AI governance remain those most affected but least resourced to participate. This includes many countries in the Global South, especially states outside major regulatory and technical blocs; your submission notes that 118 countries are absent from major AI governance discussions. It also includes Africa, Latin America, Pacific and small island states, as well as indigenous communities, rural communities, minority-language speakers, women and girls, persons with disabilities, workers affected by automation, youth, and public-interest civil society and academic researchers. These groups are often missing not because they lack stakes, but because they face structural barriers in access to funding, infrastructure, data, technical expertise and agenda-setting power. Their underrepresentation matters because AI systems are already affecting access to public services, work, language, culture, privacy and political expression. When governance debates are dominated by technologically advanced actors, global norms risk reflecting the interests of developers and major powers rather than the realities of affected communities. Your submission also highlights that rural and indigenous communities are often invisible in training datasets, while minority and marginalised voices are especially vulnerable to exclusion and harmful content moderation outcomes. They can be included through institutional design, not only invitations. The Dialogue should establish a permanent multi-stakeholder advisory mechanism with binding representation from civil society, academia, the technical community and affected communities, particularly from the Global South. It should make written submissions accessible in all six UN languages, mandate gender balance in panels and leadership, fund participation for low-resource stakeholders, and create dedicated participation channels for indigenous communities and other underrepresented groups. Inter-sessional working groups would also help ensure these voices shape outcomes continuously, not only during annual meetings.

To foster meaningful and dynamic engagement, the AI Dialogue should use formats that move participants from statement-making to problem-solving. A traditional plenary alone will not be enough. I would recommend a layered format combining political visibility with structured interaction. First, alongside the high-level plenary, the Dialogue should convene small thematic roundtables with balanced representation from governments, international organisations, the scientific community, private sector actors, civil society and affected communities. These sessions should be designed to produce short negotiated outputs: options papers, draft principles, or areas of convergence and disagreement. Second, the Dialogue should include science-policy exchange sessions in which Heads of Delegation engage directly with the Independent International Scientific Panel. Rather than one-way presentations, these should be moderated dialogues focused on evidence, trade-offs and policy implications. Your submission specifically argues for stronger interaction between scientific evidence and policy deliberation. Third, the Dialogue could use scenario-based workshops and case simulations around real policy challenges, such as AI incidents in health, education, elections or social protection. These formats help participants test how principles on safety, transparency, human rights and oversight would work in practice. Fourth, the process should remain open through written submissions, multilingual digital consultations, and inter-sessional working groups that continue engagement between annual meetings. This would prevent momentum from being lost and allow stakeholders to shape draft commitments before formal sessions. Finally, a public implementation scoreboard and commitment-review session at each Dialogue would make engagement more purposeful. The most innovative format is one where participation is not symbolic, but visibly influences decisions, follow-up and accountability.

Examples already exist across different layers of governance. A strong response is to build on what is working rather than start from zero. A useful policy model is the EU AI Act, which applies a risk-based approach: it bans certain unacceptable uses, imposes obligations on high-risk systems, and phases in rules for general-purpose AI models. That is valuable because it links regulatory obligations to the severity of risk rather than treating all AI alike. A practical public-sector governance tool is Canada's Algorithmic Impact Assessment, which is mandatory under its Directive on Automated Decision-Making. It requires departments to assess impact levels, mitigation measures, transparency and recourse before deploying automated decision systems. A strong organizational practice is the NIST AI Risk Management Framework, which gives institutions a structured way to map, measure, manage and govern AI risks. Its value lies in making risk management operational across the AI lifecycle, including generative AI. A useful capacity-building platform is UNESCO's Readiness Assessment Methodology (RAM), which helps countries identify legal, institutional and technical gaps in their preparedness to govern AI ethically and responsibly. This is especially important for countries still building baseline governance capacity. A strong rights-based legal approach is the Council of Europe Framework Convention on AI, which anchors AI governance in human rights, democracy and the rule of law, while requiring transparency, oversight and accountability measures proportionate to risk. Finally, a promising assurance mechanism is Singapore's AI Verify and its Global AI Assurance Sandbox, which translate governance principles into testing tools and real-world evaluation. That is the kind of concrete, implementation-focused approach global cooperation needs.