Digital Agency of Japan government

First, it would establish a shared baseline of understanding across countries with different levels of AI maturity. This includes agreement on core principles, such as safety, accountability, transparency, and human oversight, while recognizing diverse national contexts. Success here is not full alignment, but a common language that enables continued cooperation. Second, it would move beyond principles to practical, actionable pathways. This means identifying a small number of priority areas, such as risk management frameworks, public sector AI deployment, and data governance, and outlining voluntary, interoperable approaches. For example, agreement on minimum expectations for AI risk assessment or procurement standards in government would signal real progress. Third, it would create mechanisms for sustained collaboration. A one-off dialogue has limited value unless it leads to ongoing exchange. Establishing working groups, knowledge-sharing platforms, or pilot collaborations, particularly between advanced and emerging economies, would ensure continuity and mutual learning. Importantly, success should also be measured by inclusivity and trust. The dialogue should meaningfully incorporate perspectives from the Global South, civil society, and technical communities, not only major economies. Building trust requires openness about both opportunities and risks, as well as a willingness to learn from different governance models. Ultimately, the first dialogue should not aim to resolve all tensions, but to set a credible foundation: a shared direction, initial practical steps, and a commitment to continue working together.

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

Our selection reflects a focus on translating AI governance into practical implementation, particularly in the public sector. Safe, secure and trustworthy AI is a foundational priority. As governments increasingly deploy AI in service delivery and decision-making, ensuring safety, reliability, and risk management is essential to maintain public trust and avoid unintended harm. AI capacity-building is critical to enable effective and equitable adoption. In our context, variations in institutional capacity, across ministries and especially among local governments, significantly affect the quality and sustainability of AI deployment. Strengthening human resources, governance structures, and technical capabilities is therefore a prerequisite for meaningful implementation. Social, economic, ethical, cultural, linguistic and technical implications of AI are central to ensuring that AI systems are aligned with societal values. In particular, attention to linguistic and cultural contexts is important for inclusive AI, while ethical and economic considerations shape long-term public acceptance and impact. Transparency, accountability, and human oversight underpin responsible use. As AI systems become more embedded in administrative processes, mechanisms for explainability, auditability, and clear lines of responsibility are necessary to ensure democratic legitimacy and safeguard citizens' rights. Together, these priorities reflect a commitment to advancing AI not only as a technological capability, but as a trusted and socially grounded component of public governance.

While the listed themes capture key dimensions of AI governance, several cross-cutting and emerging issues warrant greater attention. First, AI readiness as a systemic concept remains underemphasized. Effective AI adoption depends not only on technology, but on institutional capacity, data infrastructure, governance arrangements, and workforce skills. Without aligning these elements, many AI initiatives remain at pilot stage and fail to scale,particularly in the public sector. Second, the rise of agentic AI systems introduces new governance challenges. Unlike traditional AI, these systems can autonomously plan and execute actions, increasing both their potential impact and associated risks. This shift requires rethinking accountability, control mechanisms, and safety frameworks, especially when actions, not just information, are delegated to AI. Third, there is a growing need to address the gap between global principles and local implementation. While high-level norms are increasingly converging, their translation into operational practices varies significantly across countries and institutions. Mechanisms to bridge this "implementation gap," including practical guidance, peer learning, and adaptable frameworks, are critical. Finally, trust and social acceptability should be treated as dynamic, context-dependent factors rather than static outcomes. Public confidence in AI is shaped by transparency, perceived fairness, and lived experiences of service delivery. Continuous engagement with citizens and iterative validation of AI systems are therefore essential. Addressing these cross-cutting issues would strengthen the effectiveness of AI governance by linking strategic intent with operational reality, and by ensuring that technological advances translate into trusted public value.

In our context, governance gaps across the selected thematic areas are shaping both the pace and quality of AI adoption, particularly in the public sector. A key challenge lies in uneven AI capacity across institutions. While disparities between central and local governments remain significant, it is also important to note that even within the central government, the pace and depth of AI implementation often lag behind the private sector. There has been a tendency to focus on the introduction of generative AI tools (e.g., ChatGPT-type applications), without sufficiently integrating them into broader administrative reform or institutional redesign. In parallel, competition among ministries over AI-related initiatives and unclear allocation of roles have limited the emergence of strong, coordinated leadership, leading to fragmentation and inefficiencies. In terms of safe, secure and trustworthy AI, rapid experimentation has outpaced the development of standardized risk management practices. While guidelines are emerging, consistent operationalization, such as risk assessment, monitoring, and incident response, remains underdeveloped. The social and cultural implications of AI are also increasingly visible. In a context marked by linguistic complexity and demographic challenges, ensuring inclusive and accessible AI systems is essential, alongside maintaining public trust. Regarding transparency, accountability, and human oversight, the integration of AI into administrative processes raises questions about explainability, auditability, and responsibility. Clarifying the boundaries between human and machine decision-making remains a critical issue. At the same time, these challenges present opportunities to strengthen AI readiness, develop scalable governance models, and position our approach as a practical reference for other countries.

First, it can serve as a platform for building shared understanding and trust. Given the diversity in national contexts, levels of AI maturity, and institutional capacities, a one-size-fits-all approach is neither feasible nor desirable. The Dialogue can help establish a common language around core principles, such as safety, accountability, and transparency, while fostering mutual trust among participants. Second, the Dialogue can act as a bridge from principles to implementation. While global discussions have increasingly converged around high-level principles, significant gaps remain in translating these into operational practices. By focusing on practical areas such as risk management, public procurement, and audit mechanisms, the Dialogue can facilitate the exchange of implementable approaches and promote interoperability across governance frameworks. Third, it can contribute to sustained and inclusive collaboration. Moving beyond a one-off event, the Dialogue can support the creation of working groups, pilot initiatives, and knowledge-sharing platforms that enable continuous learning. In particular, it can strengthen cooperation between advanced and emerging economies through capacity-building and joint experimentation. Importantly, the Dialogue also has a role in ensuring inclusivity. Incorporating perspectives from the Global South, civil society, and technical communities is essential for developing governance approaches that are both legitimate and adaptable. Ultimately, the AI Dialogue should not only align principles, but also enable practical cooperation that translates into trusted and effective governance outcomes.

The AI Dialogue should build upon and connect with existing international initiatives that have already advanced principles, measurement frameworks, and practical cooperation in AI governance. Key among these are the OECD AI Principles and related implementation tools, which provide a widely endorsed foundation for trustworthy AI. The G7 Hiroshima AI Process has further contributed to international norm-setting, particularly in the context of generative AI, by promoting risk-based approaches and voluntary codes of conduct. In addition, the Global Partnership on AI (GPAI) offers a platform for applied collaboration, bringing together governments, academia, and industry to develop practical insights. Efforts within the United Nations system, including discussions under the General Assembly and specialized agencies, also play an important role in ensuring global inclusivity and legitimacy. The added value of the AI Dialogue lies not in duplicating these efforts, but in connecting and operationalizing them. Specifically, the Dialogue can serve as a convening platform to align these initiatives, reduce fragmentation, and promote coherence across different governance approaches. It can also focus on bridging the gap between high-level principles and real-world implementation by facilitating the exchange of concrete practices, particularly in public sector use cases. Furthermore, the Dialogue can provide a space to address cross-cutting issues, such as AI readiness, capacity-building, and interoperability, that span across existing initiatives but are not always treated in an integrated manner. By building on established foundations while emphasizing coordination and implementation, the AI Dialogue can enhance the overall effectiveness and impact of global AI governance efforts

To enhance the effectiveness of the AI Dialogue, it is essential for diverse stakeholders to contribute based on their respective strengths within a clearly defined structure. Governments can lead by sharing policy frameworks, regulatory approaches, and practical experiences from public sector implementation. The private sector can contribute cutting-edge technical expertise and real-world use cases, offering insights into feasibility and scalability. Academia plays a key role in providing independent analysis, evaluation, and evidence, ensuring rigor and neutrality in the discussion. Civil society brings critical perspectives on inclusivity, human rights, and social acceptability. To maximize these contributions, the format and structure of the AI Dialogue should be carefully designed. First, in addition to plenary sessions, thematic working groups should be established to enable sustained and focused discussions on specific issues. Second, practical sessions centered on pilot projects and case studies should be incorporated to bridge the gap between principles and implementation. Third, regular reporting and sharing of outcomes are important to ensure transparency and accountability. Furthermore, adopting a hybrid format that combines in-person and virtual participation would allow for broader engagement beyond geographical constraints, particularly enabling participation from the Global South. With such a structure, the AI Dialogue can function as an inclusive and practice-oriented platform for international cooperation on AI governance.

Several important voices and perspectives remain underrepresented in current global discussions on AI governance. First, governments and practitioners in the Global South and across Asia are often underrepresented. Asia is a region where AI adoption is rapidly advancing, yet its diverse institutional settings and social contexts are not sufficiently reflected in global discussions. At the same time, the experiences of the Global South in implementing AI under resource constraints remain undervalued. Second, local governments and frontline public service practitioners are rarely included, despite being directly responsible for delivering services where the impact of AI is most tangible. Third, linguistically and culturally diverse communities, particularly non-English-speaking populations, are insufficiently reflected, limiting the inclusiveness and relevance of AI systems. Fourth, affected communities and vulnerable groups are often not meaningfully engaged, even though their perspectives are essential for addressing fairness, equity, and social acceptance. To better include these voices, mechanisms beyond formal representation are needed. This includes organizing regional and thematic dialogues, creating dedicated forums for practitioners, and strengthening multilingual engagement. Reducing participation barriers, such as cost, time, and access constraints, through financial support and expanded virtual participation is also critical. In addition, integrating real-world experiences through pilot projects and case studies can help ground discussions in practical realities.

To foster meaningful and dynamic engagement in the AI Dialogue, formats should move beyond traditional presentation-based approaches and be designed to actively harness participants' knowledge. First, collective intelligence approaches should be leveraged. AI-enabled tools can be used to aggregate and visualize diverse inputs in real time, enabling structured insights to emerge beyond simple exchanges of views. Second, gamification can enhance engagement. Introducing game-based elements into problem-solving processes can encourage active participation and deepen understanding of complex policy challenges. For example, competitive policy design exercises or role-playing simulations across different stakeholder perspectives can be effective. Third, strategic foresight should be integrated. Through scenario-based discussions and backcasting, participants can explore long-term uncertainties and move beyond short-term thinking, enabling more resilient and forward-looking governance approaches. These methods should be complemented by use-case driven sessions and co-creation labs, which help connect principles with implementation. In addition, establishing continuous engagement through hybrid platforms, combining in-person and virtual participation, can ensure sustained interaction and knowledge accumulation. By integrating these approaches, the AI Dialogue can evolve into a practical platform that simultaneously supports knowledge creation, policy innovation, and collaborative problem-solving.

First, the deployment of generative AI within government operations alongside the development of usage guidelines provides a practical approach to governance. By using AI tools in real workflows, rules and risk management practices are developed iteratively based on real experience. Second, the appointment of Chief AI Officers (CAIOs) across ministries helps clarify accountability for both AI adoption and risk management, while accelerating decision-making within each organization. Third, initiatives on AI social readiness for citizens are an important component. For example, in collaboration with Nesta, pilot programs have been conducted to explore how citizens understand and perceive AI. Such initiatives help strengthen social acceptance and incorporate public perspectives into policy design. Fourth, collaborative research with universities on AI ethics and algorithms provides an additional governance mechanism. In particular, working with university students to conduct surveys and studies on AI ethics and algorithmic impacts helps integrate the perspectives of future users and stakeholders into policymaking. These practices reflect an approach in which implementation and governance evolve together through iterative learning and adaptation.