Loughborough University

A successful first Global Dialogue on AI Governance would, in my view, produce three things. First, it should establish a genuinely shared international baseline for AI governance. That does not mean forcing every country into identical rules, but agreeing a common core of principles around safety, transparency, accountability, human rights, inclusion, and public benefit. Success would be a clear sense that governments, industry, academia, and civil society are moving toward a common direction rather than a fragmented patchwork of competing approaches. Second, it should broaden the governance conversation beyond model behaviour alone to include the full AI system lifecycle. A strong outcome would be recognition that AI governance must also address data practices, infrastructure demands, environmental impacts, and resource use. In particular, the Dialogue would be stronger if it acknowledged that AI has implications not only for ethics and security, but also for energy demand, water use, digital waste, and the growing carbon burden associated with storage, network transfer, and compute. That would help position sustainability as a core governance issue rather than a peripheral one. Third, it should end with a practical roadmap rather than a general statement of intent. Success would be a set of concrete next steps: priority themes for international cooperation, areas where standards or metrics are needed, mechanisms for knowledge sharing, and a commitment to include voices from lower-income countries and underrepresented regions in shaping future governance. Ultimately, the first Dialogue will be successful if it creates momentum, legitimacy, and usable direction. It should make participants feel that global AI governance is not simply a diplomatic discussion, but the beginning of a credible, inclusive, and action-oriented process.

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

1. Safe, secure and trustworthy AI This is foundational. Public confidence in AI depends on systems being reliable, robust, and governed in ways that reduce harm. For our work, trustworthiness also extends beyond model outputs to the wider system conditions under which AI is built and deployed. 2. Transparency, accountability, and human oversight These are essential if AI governance is to be meaningful in practice. Organisations need greater visibility over what AI systems do, what data they rely on, who is responsible for decisions, and what oversight mechanisms are in place. This is also important for understanding the wider impacts of AI systems, including operational, environmental, and societal consequences. 3. Social, economic, ethical, cultural, linguistic and technical implications of AI This category best reflects the breadth of real-world AI impacts. It creates space to discuss not only fairness and social outcomes, but also infrastructure pressures, resource consumption, and unintended externalities. From our perspective, it is especially important that these implications include the environmental effects of AI across data, storage, network, and compute. 4. AI capacity-building Capacity-building is critical because effective governance depends on people and institutions being able to understand, evaluate, and manage AI responsibly. This includes technical capability, policy capability, and organisational capability. It should also include building awareness of efficient and sustainable AI practices, so that adoption does not unintentionally increase digital waste, energy demand, and avoidable emissions. Together, these four priorities balance immediate governance needs with longer-term systemic resilience. They combine the need for trustworthy and accountable systems with the need to build capability and to address the full range of AI's implications, including sustainability.

One important cross-cutting issue that is not yet sufficiently visible in the listed themes is the environmental sustainability of AI across the full digital lifecycle. Current AI governance discussions often focus, rightly, on safety, rights, transparency, and accountability. However, there is a growing need to recognise that AI systems also create material environmental pressures through the volume of data they require, the infrastructure they depend on, and the resources consumed across storage, network transfer, and compute. Governance discussions that focus only on model behaviour risk overlooking these broader system-level impacts. This matters because AI can increase energy demand, water use, hardware dependency, and digital waste, particularly where large volumes of data are stored but never reused, or where models are developed and deployed without sufficient attention to efficiency. In practice, this means that AI governance should not only ask whether a system is fair, safe, or explainable, but also whether it is proportionate, efficient, and environmentally responsible by design. A useful emerging concept here is digital decarbonisation: reducing the environmental impact of digital systems by improving how data is created, stored, moved, processed, and governed. Bringing this perspective into the AI Dialogue would help link governance to sustainability, infrastructure planning, and responsible innovation. This is not a separate concern from the listed themes; it cuts across them. It relates to trustworthy AI, transparency, capacity-building, and the social and economic implications of AI. Recognising environmental sustainability as a cross-cutting issue would strengthen the Dialogue by ensuring that global AI governance addresses not only what AI does, but also the resources it consumes and the long-term burdens it creates.

In our sector, the main governance gap is that AI discussion is moving faster than the frameworks needed to guide responsible deployment in practice. This creates both strategic risk and missed opportunity. The most significant challenge is fragmentation. Organisations are being asked to adopt AI in ways that are safe, transparent, and economically beneficial, but there is still limited clarity on what good governance looks like across the full system lifecycle. In practice, this means uneven approaches to risk management, accountability, data quality, oversight, and measurement. It also means that important impacts can be overlooked, particularly the wider infrastructure consequences of AI, including rising demand for data storage, network capacity, compute power, energy, and cooling. A further challenge is that governance still tends to focus on model outputs rather than system inputs and operating conditions. In our view, this leaves a gap around efficiency, proportionality, and environmental sustainability. Without clearer guidance, organisations may scale AI in ways that increase cost, digital waste, and carbon emissions, even where the underlying value is uncertain. At the same time, there is a major opportunity. Better governance can help organisations adopt AI with greater confidence, improve accountability, and support innovation that is both effective and responsible. There is also an opportunity to embed sustainability earlier in the governance agenda by encouraging more efficient data practices, greater transparency over resource use, and stronger attention to storage, network, and compute impacts. For our region and sector, the greatest benefit would come from governance approaches that are practical, internationally interoperable, and broad enough to include not only ethics and safety, but also the environmental footprint of AI systems. That would support more trustworthy adoption while helping avoid avoidable digital waste and long-term infrastructure burdens.

The AI Dialogue can play an important role by creating a credible international space for alignment, learning, and practical cooperation on AI governance. Its value should not be in producing abstract consensus alone, but in helping countries and sectors move towards more compatible, actionable, and inclusive approaches. First, it can help build a shared understanding of what responsible AI governance should cover. This includes not only safety, security, transparency, accountability, and human rights, but also the wider system conditions in which AI is developed and deployed. A valuable contribution of the Dialogue would be to encourage a more holistic view of governance across the AI lifecycle, from data and model development to deployment, oversight, and infrastructure impacts. Second, it can support cooperation on standards, metrics, and good practice. Many organisations and governments are facing similar questions but addressing them in isolation. The Dialogue could help identify areas where international coordination is most needed, including risk assessment, transparency mechanisms, capacity-building, and methods for measuring AI's broader impacts. Third, it can strengthen cooperation by bringing emerging issues into the mainstream of governance discussions. One area where this would be especially useful is the environmental sustainability of AI. As AI adoption grows, so too do pressures linked to data generation, storage, network transfer, compute demand, energy use, cooling, and digital waste. The Dialogue could help ensure that these issues are treated as part of responsible AI governance rather than as a separate technical concern. In this respect, digital decarbonisation offers a useful lens for international cooperation, by encouraging more efficient, proportionate, and sustainable AI systems by design. Overall, the Dialogue can add value by making international cooperation more practical, more inclusive, and more forward-looking. It can help ensure that global AI governance addresses not only what AI systems do, but also how they are built, resourced, and sustained.

The AI Dialogue should build on initiatives that already provide recognised principles, implementation tools, and convening power. In particular, it should connect with the UN Global Digital Compact, which explicitly frames digital cooperation and AI governance as part of the wider multilateral agenda; UNESCO's Recommendation on the Ethics of AI, which applies to all 194 UNESCO Member States; and the OECD AI Principles, updated in 2024, together with the integrated GPAI/OECD.AI partnership, which provides policy guidance, shared learning, and practical implementation support. It should also connect with existing standards and operational frameworks, including ISO/IEC JTC 1/SC 42 on AI standardisation and the NIST AI Risk Management Framework, which offers a practical structure for governing, mapping, measuring, and managing AI risks. In addition, the UN Secretary-General's Governing AI for Humanity report and the AI for Good platform provide useful foundations for a more coordinated global architecture, and AI for Good 2026 is already being run back-to-back with the Global Dialogue in Geneva. The added value of the AI Dialogue should be to connect these efforts politically and internationally, rather than duplicate them. It can help bridge principles, standards, and practice across regions; identify priority gaps where interoperability or capacity-building is most needed; and ensure that voices beyond the main regulatory and technical centres are included. It can also add value by bringing in cross-cutting issues that are often underplayed across existing initiatives, especially the environmental sustainability of AI. From a digital decarbonisation perspective, that means giving greater attention to the impacts of data growth, storage, network transfer, compute demand, energy use, cooling, and digital waste across the AI lifecycle.

Different stakeholders should contribute in ways that reflect their distinct expertise, responsibilities, and lived realities, rather than through a single uniform process. Governments can provide policy direction and identify where international alignment is most needed. Industry can share implementation experience, operational constraints, and emerging practice. Academia can contribute evidence, independent evaluation, and longer-term analysis. Civil society can help ensure that public interest, inclusion, rights, and real-world impacts remain central. Technical standards bodies and professional associations can help translate broad principles into usable methods, metrics, and guidance. One useful model is the way a Digital Decarbonisation Strategic Advisory Group (DD SAG) can operate: bringing together stakeholders from policy, industry, research, and practice in a structured but practical forum focused not only on discussion, but on surfacing shared priorities, areas of disagreement, and actionable next steps. That kind of model is valuable because it allows strategic issues to be examined from multiple perspectives while still moving towards outputs that can inform practice and policy. In terms of format, the AI Dialogue would benefit from a structure that combines high-level plenary discussion with smaller thematic working sessions. Plenaries are useful for shared vision and political visibility, but smaller roundtables are better for detailed exchange, problem solving, and identifying concrete areas for cooperation. A good structure might include: a plenary to set priorities; themed breakout sessions with balanced stakeholder participation; and a final synthesis session focused on practical recommendations and follow-up. The Dialogue should also be designed as a continuing process, not a one-off event. Its value will be greatest if it creates mechanisms for ongoing engagement between annual meetings, such as expert working groups, evidence submissions, and shared outputs. That would help ensure that diverse stakeholders do not simply attend the Dialogue, but actively shape its direction and usefulness over time.

The most underrepresented voices in global AI governance are often the people who live with the consequences of AI but have the least power to shape how it is designed, deployed, or governed. In practice, global discussions are still heavily influenced by governments, major technology firms, and a relatively small group of well-resourced institutions. As a result, the perspective of the end user is often weakly represented. This matters because AI is already creating a new kind of digital divide. For example, in universities some students can afford paid large language model services, while others rely on free versions with more limited capabilities. Over time, differences in access to AI tools, model quality, speed, and functionality may begin to influence learning opportunities, productivity, and even degree outcomes. The same pattern is likely to appear more broadly across workplaces, schools, and public services. Those with access to better AI may move ahead faster, while others are left behind. Other underrepresented groups include workers using AI in practice, SMEs without the compliance capacity of large firms, low- and middle-income countries, minority-language communities, civil society organisations with limited funding, and those concerned with the environmental and infrastructure impacts of AI. These groups will not be included unless participation is designed deliberately. The Dialogue should provide funded access, remote participation, interpretation, advance evidence channels, and smaller facilitated discussions that do not privilege only the loudest or best-resourced actors. A good test is simple: global AI governance should not be shaped only by those building and selling AI, but also by those using it, depending on it, and at risk of being excluded by it.

Meaningful engagement during the AI Dialogue will depend less on novelty alone and more on formats that are effective, inclusive, and credible. The priority should be to create settings where diverse stakeholders can contribute substantively, rather than simply make brief set-piece interventions. A strong format would combine high-level plenary sessions with smaller facilitated working dialogues. Plenaries are important for setting shared direction and political visibility, but smaller sessions are more effective for testing ideas, surfacing disagreement, and identifying practical areas for cooperation. These should be carefully moderated to ensure discussion is not dominated by the most powerful or best-resourced participants. It would also be valuable to include multi-stakeholder roundtables organised around specific governance problems, rather than only broad themes. For example, sessions could focus on transparency in practice, AI access and inequality, public sector use, or the environmental sustainability of AI systems. This allows governments, industry, academia, civil society, and end users to engage around concrete questions and produce more actionable insights. Another useful format would be evidence-led dialogues, where short framing inputs are circulated in advance and discussions are structured around responding to them. This would help move the Dialogue beyond general statements towards more grounded exchange. Finally, the process should include mechanisms for participation beyond the room: written submissions, remote engagement, moderated digital participation, and structured synthesis of inputs from underrepresented groups. Overall, the best engagement formats are those that combine openness with discipline: broad enough to hear different perspectives, but structured enough to produce clear lessons, priorities, and follow-up actions.

Effective AI governance is most likely to emerge when agenda-setting platforms and practical policy frameworks work together. In that respect, two useful examples are the World Economic Forum's white papers and multi-stakeholder alliances, which help surface emerging risks, opportunities, and business-relevant governance questions, and the OECD AI Principles and OECD.AI/GPAI architecture, which help translate these into policy guidance that governments can adopt and adapt. The OECD AI Principles were updated in 2024 and remain one of the clearest international reference points for trustworthy AI, while the integrated OECD/GPAI model provides an ongoing mechanism linking governments with expert communities. A strong example of useful practice is therefore an approach that connects strategic thought leadership with implementable public policy. The Forum can help frame where attention is urgently needed, convene industry and cross-sector actors, and test emerging governance ideas. The OECD and related intergovernmental mechanisms can then help convert these insights into more durable principles, policy tools, and internationally comparable approaches. This matters especially for issues that are still underrepresented in mainstream governance discussions, such as the environmental sustainability of AI. Recent WEF work on AI and energy has helped bring attention to the growing infrastructure consequences of AI adoption, including electricity demand and the need for more efficient deployment. The added value of combining this with policy-oriented mechanisms is that governments are more likely to act when emerging concerns are linked to recognised frameworks, standards, and implementation pathways. In short, effective AI governance needs both: forums that help shape direction and policy mechanisms that help operationalise it. The AI Dialogue could add value by acting as the bridge between them.