The Centre for Sustainable AI

A successful first Global Dialogue on AI Governance would produce outcomes that are practical, inclusive, and capable of building long-term trust across countries, sectors, and communities. First, success would mean establishing a shared baseline of principles for safe, responsible, and human-centred AI. This does not require full global consensus on every issue, but it should create common ground on core matters such as transparency, accountability, safety, fairness, privacy, and human oversight. Second, the Dialogue should help move the conversation from high-level principles to implementation. A meaningful outcome would be agreement on priority areas for action, such as risk classification, testing and assurance, governance of frontier and agentic AI, incident reporting, and mechanisms for monitoring real-world impacts. Even a simple roadmap for collaborative work would be valuable. Third, success would require genuine inclusion. The Dialogue should amplify the voices of developing countries, small states, Indigenous communities, civil society, academia, and industry, rather than allowing governance to be shaped only by major powers or large technology companies. AI governance will only be legitimate if it reflects diverse social, economic, and cultural realities. Fourth, it should build international cooperation rather than fragmentation. This could include commitments to knowledge sharing, regulatory interoperability, capacity building, and support for countries that are still developing governance capability. Finally, the strongest sign of success would be momentum beyond the event itself. If the Dialogue leads to ongoing working groups, measurable follow-up actions, and a trusted platform for continued collaboration, then it will have achieved more than discussion. It will have laid the foundation for a credible global governance ecosystem that supports innovation while protecting humanity.

• Safe, secure and trustworthy AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

Our priorities reflect the need to ensure that AI is developed and deployed in ways that are safe, trustworthy, and centred on human wellbeing. Safe, secure and trustworthy AI is essential because public confidence, resilience, and long-term adoption depend on systems that are reliable, robust, and protected against misuse. We also prioritise the social, economic, ethical, cultural, linguistic and technical implications of AI because AI does not operate in a vacuum. Its impacts are shaped by context, and governance must account for how AI may affect different communities, languages, cultures, and levels of economic development. Without this broader perspective, governance risks being technically narrow and socially incomplete. The protection and promotion of human rights is another urgent priority. AI governance should uphold dignity, equality, privacy, non-discrimination, and access to remedy. Human rights provide a universal foundation that can help guide AI development across different national and institutional settings. Finally, transparency, accountability, and human oversight are critical for operationalising trust. People and institutions must be able to understand how AI is used, who is responsible for decisions and outcomes, and where meaningful human review and intervention remain necessary. These elements are essential for assurance, contestability, and responsible innovation. Taken together, these four priorities provide a practical and principled foundation for global AI governance that supports innovation while safeguarding people and society.

Yes. While the listed themes cover many core areas, several cross-cutting and emerging issues deserve more explicit attention. One is AI assurance and auditability. It is not enough to define principles such as safety, transparency, and accountability. There must also be practical mechanisms to test, verify, monitor, and independently audit AI systems across their lifecycle, including before deployment and during ongoing use. A second is governance of frontier and agentic AI systems. Increasingly autonomous systems raise new questions around control, escalation, delegation of decision-making, tool use, memory, and safe boundaries for action. These issues cut across safety, human oversight, and accountability, but are significant enough to warrant distinct attention. A third is compute, infrastructure, and concentration of power. AI governance is not only about models and applications. It is also about who controls the data, compute, cloud infrastructure, and deployment ecosystems that shape access, competition, and global equity. This has implications for inclusion, sovereignty, and dependency. A fourth issue is environmental sustainability. The energy, water, and material costs of large-scale AI systems should be more explicitly recognised within global governance discussions, especially as AI adoption expands. Finally, regulatory interoperability and implementation capacity remain important cross-cutting concerns. Many countries and organisations do not lack principles. They lack the institutional capability, technical expertise, and practical tools to implement them consistently.

In Australia and the broader Indo-Pacific, governance gaps in safe, trustworthy, and human-centred AI are creating both immediate risks and important opportunities. The most significant challenge is the gap between high-level principles and operational implementation. Many organisations recognise the need for safety, accountability, human oversight, and rights protection, but still lack consistent assurance processes, clear accountability structures, technical testing capabilities, and governance maturity across the full AI lifecycle. This is particularly important in sectors such as telecommunications, critical infrastructure, government services, health, and education, where AI can influence access, safety, service quality, and public trust at scale. Governance gaps can lead to opaque decision-making, bias, privacy risks, weak procurement oversight, and limited ability to monitor third-party or rapidly evolving generative and agentic AI systems. In our region, uneven regulatory readiness and capability across countries also creates fragmentation, making cross-border interoperability and trusted collaboration more difficult. At the same time, these developments present a major opportunity. Stronger governance can position Australia and the region as leaders in trusted AI adoption. There is a significant opportunity to build practical assurance ecosystems around standards, audits, incident reporting, risk classification, and human oversight. There is also an opportunity to invest in capacity-building so that smaller organisations, public institutions, and developing economies are not left behind. Most importantly, good governance can enable innovation rather than slow it. When AI systems are designed with safety, transparency, accountability, and human rights in mind, organisations are better placed to deploy them with confidence, earn public trust, and achieve sustainable social and economic value.

The AI Dialogue can play a valuable role as a neutral, inclusive, and action-oriented platform for building international cooperation on AI governance. Its greatest contribution would be to help bridge the current gap between fragmented national, regional, and sectoral initiatives and the need for more coherent global coordination. First, the Dialogue can support the development of shared understanding. Countries and stakeholders are approaching AI governance from different legal, economic, and institutional contexts, but there is still strong value in identifying common principles, shared terminology, and areas of practical convergence. This can improve trust and reduce unnecessary fragmentation. Second, it can promote regulatory and governance interoperability. The goal is not to impose a single global model, but to encourage alignment across frameworks, standards, assurance approaches, and risk management practices so that cooperation becomes more practical across borders. Third, the Dialogue can strengthen inclusion and capacity-building. Many countries, particularly developing economies and small states, need greater access to technical expertise, governance tools, and policy support. The Dialogue can help ensure that international cooperation is not shaped only by the most technologically advanced states or largest companies. Fourth, it can create pathways for collaboration on emerging issues such as frontier AI, agentic systems, safety testing, incident reporting, and cross-border accountability. These are areas where isolated national action will be insufficient. Finally, the Dialogue can help sustain momentum through follow-up mechanisms such as working groups, knowledge-sharing networks, and collaborative implementation initiatives. Its value will depend not only on discussion, but on whether it enables ongoing cooperation, practical exchange, and measurable progress.

The AI Dialogue should build on initiatives that already provide substantive foundations for global AI governance. These include UNESCO's Recommendation on the Ethics of AI, which gives a global normative baseline across all Member States; the OECD AI Principles and the integrated OECD–GPAI partnership, which provide policy guidance, implementation tools, and international policy cooperation; the G7 Hiroshima AI Process, which has advanced work on governance for advanced and generative AI; and the Council of Europe Framework Convention on AI, which is the first legally binding international treaty in this space. It should also connect with ISO/IEC JTC 1/SC 42 standards work, and with ITU-led platforms such as AI for Good and WSIS, which bring technical, standards, and capacity-building communities into the conversation. Within the UN system, the Dialogue should also align with the Global Digital Compact, the UN Secretary-General's AI Advisory Body recommendations, and the new Independent International Scientific Panel on AI, so that political dialogue is informed by evidence-based assessment and does not duplicate parallel processes. Its added value would be legitimacy, inclusiveness, and coordination. Unlike narrower forums, the AI Dialogue can provide a universal multilateral space where developing countries, smaller states, civil society, academia, and industry all have a meaningful voice. It can help connect fragmented initiatives, promote interoperability across principles, standards, and regulatory approaches, and turn broad consensus into practical follow-up on issues such as assurance, capacity-building, safety evaluation, and cross-border accountability. In that sense, the Dialogue should not replace existing mechanisms, but serve as the convening layer that links them into a more coherent global governance ecosystem.

Different stakeholders should contribute in ways that reflect their distinct responsibilities and expertise. Governments can share policy experience, regulatory needs, and public-interest priorities. Industry can contribute technical knowledge, deployment experience, and lessons from implementation. Academia and independent researchers can provide evidence, foresight, and evaluation methods. Civil society, affected communities, workers, and human rights advocates are essential for ensuring that governance reflects lived impacts, not only institutional or commercial perspectives. International organisations and standards bodies can help connect the Dialogue to existing frameworks, norms, and capacity-building efforts. For the Dialogue to be effective, participation should be meaningful rather than symbolic. This means balanced representation across regions, especially from developing countries, small states, and underrepresented communities, with sufficient support for participation. In terms of format, the Dialogue should combine high-level political engagement with practical working-level collaboration. A useful structure would include a plenary forum to set shared priorities, followed by thematic working groups focused on issues such as safety and assurance, human rights, frontier and agentic AI, interoperability, and capacity-building. Regional and sector-specific consultations should feed into the global process so that outcomes reflect different contexts and levels of readiness. The Dialogue should also be evidence-based and action-oriented. Background papers, case studies, and technical briefings should inform discussions in advance. Each session should aim for concrete outputs such as areas of convergence, priority actions, and follow-up commitments. Finally, the Dialogue should not be a one-off event. It should include a light but credible continuity mechanism, such as recurring working groups, regular progress reviews, and a shared platform for knowledge exchange. That is how multistakeholder participation can translate into practical international cooperation.

Several voices remain underrepresented in global AI governance discussions. Most notably, developing countries, small states, and communities in the Global South are often not included on equal terms, even though they will be significantly affected by AI adoption, dependency on foreign technologies, and uneven governance capacity. Indigenous communities are also underrepresented, despite offering important perspectives on collective rights, stewardship, and culturally grounded approaches to data and technology. Workers, educators, children and young people, persons with disabilities, and communities most exposed to automated decision-making are also too often discussed as affected groups rather than engaged as participants. There is also a gap in representation from non-English-speaking and culturally diverse communities. Global AI governance can become too shaped by a narrow set of legal, technical, and commercial assumptions from a small number of countries and firms. These voices should be included through deliberate design, not assumption. This means funded participation support, multilingual engagement, regional consultations, accessible formats, and transparent stakeholder selection processes. It also means creating space for civil society, community leaders, labour representatives, and affected groups to contribute to agenda-setting, not only to react to draft outcomes. Inclusion should go beyond symbolic participation. Underrepresented groups should have real influence over priorities, working groups, and follow-up mechanisms. Global AI governance will only be credible, legitimate, and effective if it reflects the diversity of the societies it is meant to serve.

Meaningful engagement is more likely when the AI Dialogue moves beyond formal statements and creates settings where participants can test ideas, compare experiences, and work through practical trade-offs together. One effective format would be multistakeholder scenario labs. These sessions could use realistic case studies such as frontier AI incidents, cross-border harms, public-sector deployment, or failures in human oversight to bring governments, industry, civil society, and researchers into the same problem-solving space. This helps ground abstract principles in operational reality. A second useful format is moderated policy sprints. Small, diverse groups could work over a short period on specific questions such as assurance mechanisms, interoperability, capacity-building, or governance of agentic AI, then report back with concrete options rather than general discussion. A third format is regional and community listening forums integrated into the main Dialogue. These would allow perspectives from developing countries, Indigenous communities, youth, workers, and other underrepresented groups to shape the agenda in a structured way rather than being treated as side conversations. The Dialogue could also include expert evidence sessions that function like short technical briefings. These would help participants engage with emerging issues such as safety evaluations, compute governance, environmental impacts, and incident reporting on a common factual basis. Finally, interactive commitment and follow-up sessions would be valuable. Instead of ending with broad declarations, stakeholders could identify practical next steps, partnership opportunities, and areas for continued collaboration through working groups or shared implementation tracks.

A strong policy example is the EU AI Act, which uses a risk-based approach to match obligations to the level of risk posed by an AI system, including transparency duties and additional requirements for high-risk and general-purpose AI models. Its value is that it turns broad principles into concrete regulatory obligations. A strong practice-oriented example is the NIST AI Risk Management Framework. It helps organisations identify, assess, and manage AI risks across the lifecycle, which makes governance more operational. ISO/IEC 42001 adds further value by embedding AI governance into an organisation-wide management system with policies, processes, accountability, and continual improvement. A useful public-sector model is Canada's Directive on Automated Decision-Making, supported by its Algorithmic Impact Assessment tool. This combines risk scoring with transparency, assurance, public reporting, and review mechanisms, offering a practical example of how governments can govern automated systems in a structured way. A promising platform example is Singapore's AI Verify ecosystem. It provides testing and assurance tools that help organisations assess AI systems against recognised governance principles, making technical assurance more tangible and usable in practice. At the global level, the OECD AI Principles and the UNESCO Recommendation on the Ethics of AI remain important because they provide shared normative foundations for trustworthy, human-centred, and internationally interoperable AI governance.