2021.AI

A successful first Global Dialogue on AI Governance should produce practical outcomes that move beyond principles and toward shared implementation. In my view, success would mean three things: first, a common baseline for safe, secure and trustworthy AI that can be adapted across jurisdictions without fragmenting innovation; second, a credible commitment to closing the AI capacity gap through access to infrastructure, skills and open, reusable tools; and third, a stronger global norm that high-impact AI must remain transparent, accountable and subject to meaningful human oversight. From my perspective as a builder of enterprise AI governance and infrastructure, the Dialogue will matter most if it helps align governance with reality. Too often, the debate is framed as a choice between innovation and control. It is not. The real task is to create the conditions under which AI can be deployed at scale without compromising rights, resilience or institutional trust. That requires interoperable governance approaches, clear accountability lines, and operational standards that work across public and private sectors. A concrete success would be agreement on a shared set of minimum expectations for high-risk AI systems, alongside stronger support for developing countries to build digital foundations, access AI applications and participate in the governance of the technology they will be expected to use. If the Dialogue can help turn AI governance into something more inclusive, more implementable and more globally coherent, it will have made a real difference.

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

Our priorities for urgent action are: safe, secure and trustworthy AI; AI capacity-building; interoperability of governance approaches; and transparency, accountability, and human oversight. These areas reflect the most immediate needs if AI is to be deployed responsibly at scale. First, safe, secure and trustworthy AI is essential because organisations will only adopt AI broadly when systems are robust, resilient, and designed to perform reliably in real-world settings. Second, AI capacity-building is critical to ensure that institutions, public authorities, and smaller markets are not left behind due to limited technical, regulatory or operational capabilities. Third, interoperability of governance approaches is increasingly important. AI is a global technology, but governance is still highly fragmented across jurisdictions. If we want AI to scale across borders while remaining compliant and trusted, governance frameworks must be sufficiently aligned to allow practical implementation without creating unnecessary complexity or duplication. Finally, transparency, accountability, and human oversight are fundamental to maintaining trust in AI systems, especially in high-impact and regulated contexts. AI should support human decision-making, not replace responsibility for it. Clear lines of accountability and meaningful oversight are necessary to protect rights, reduce risk, and support adoption. Together, these priorities reflect a pragmatic view: AI governance must enable innovation, but it must do so in a way that is secure, interoperable, and grounded in human responsibility.

Yes, one important cross-cutting issue is AI sovereignty, especially the dependence on external models, cloud infrastructure, data pipelines, and vendor-controlled deployment conditions. This is not fully captured by the listed themes, but it is increasingly central to whether countries and institutions can retain meaningful control over critical AI capabilities, particularly in regulated sectors and public services. A second emerging issue is the need for operational AI governance across the full lifecycle. Many governance discussions focus on model development or initial approval, but in practice the biggest risks often emerge after deployment: drift, degradation, misuse, weak monitoring, and unclear accountability when systems are updated or embedded into complex workflows. Lifecycle governance should therefore be treated as a core issue in its own right. A third important issue is the interaction between AI and critical infrastructure resilience. As AI becomes more embedded in healthcare, energy, finance, and public administration, the question is not only whether AI is trustworthy, but whether the institutions using it are resilient against outages, cyber risk, supply-chain dependencies, and vendor concentration. Finally, I would highlight practical implementation capacity. Many organisations understand the principles of responsible AI, but lack the operational frameworks, skills, and tooling needed to apply them consistently. This gap between policy ambition and implementation reality is likely to become one of the most decisive issues in AI governance.

In Denmark and across Europe, the main governance gap is not a lack of interest in responsible AI, but the lack of operational structures that allow AI to be deployed safely, repeatedly, and at scale in regulated environments. In sectors such as healthcare, financial services, and the public sector, organisations increasingly recognise the value of AI, but adoption is slowed by uncertainty around compliance, accountability, monitoring, and procurement readiness. A significant challenge is the fragmentation of governance approaches across jurisdictions and institutions. This creates complexity for organisations operating across borders and makes it harder to scale trusted AI solutions efficiently. Another challenge is the dependency on external cloud, model, and infrastructure providers, which can weaken sovereignty and limit local control over data, deployment conditions, and long-term resilience. At the same time, these gaps create a major opportunity. Denmark has strong digital foundations, high trust in institutions, and a mature public-private collaboration culture. That gives the country a strong basis for becoming a leader in operational AI governance — especially in areas such as lifecycle management, model monitoring, and trusted deployment in regulated settings. If governance frameworks become more interoperable and implementation-focused, they can accelerate adoption rather than slow it down. For the sector I work in, the opportunity is particularly clear: organisations are looking for ways to move from pilots to production without compromising safety, accountability, or compliance. This creates demand for governance-enabled AI platforms that can support deployment across the full lifecycle and across multiple institutions. In that sense, the governance gap is also a market opportunity for European providers building sovereign, trustworthy AI infrastructure.

The AI Dialogue can play a valuable role by creating a regular global forum where governments, industry, academia and civil society can move from abstract debate to practical cooperation. Its most important contribution would be to help establish shared expectations for safe, secure and trustworthy AI while respecting different legal systems, levels of capacity and policy priorities. One key role is to reduce fragmentation. AI governance is currently developing in parallel across regions, which creates uncertainty for organisations operating internationally and raises the risk of inconsistent standards. The Dialogue can help identify common minimum principles and areas where interoperability is possible, making it easier to align approaches without forcing full harmonisation. A second role is to support capacity-building, especially for countries that lack access to infrastructure, technical expertise or regulatory resources. International cooperation on AI governance will only be credible if it helps narrow the gap between countries that are shaping AI systems and those that are mainly consumers of them. Third, the Dialogue can provide a platform for operational learning: what works in practice for transparency, accountability, human oversight, monitoring and lifecycle governance. That kind of exchange is essential if governance is to keep pace with the real-world deployment of AI in critical sectors. Ultimately, the Dialogue can help build trust between stakeholders and across regions. If it produces practical cooperation, shared language and implementable guidance, it can become an important bridge between global principles and local execution.

The AI Dialogue should build on the growing ecosystem of existing AI governance efforts rather than duplicate them. That includes the UNESCO Recommendation on the Ethics of AI, the OECD's AI work and GPAI, the EU AI Act, the Council of Europe's AI treaty process, the G7 Hiroshima Process, the UN Secretary-General's High-Level Advisory Body on AI, and UNESCO's global policy and capacity-building initiatives. Its added value would be to provide a more inclusive UN-based convening space that connects these efforts, reduces fragmentation, and helps translate principles into practical, interoperable implementation. It can also amplify capacity-building by linking low- and middle-income countries to expertise, tools, and shared learning, building on existing models such as UNESCO's "AI Ethics Experts without Borders" and related training and policy support. A further contribution would be to bridge the gap between global norms and operational reality. The Dialogue can bring together governments, technical experts, industry, and civil society to discuss lifecycle governance, accountability, human oversight, and deployment in high-impact sectors in a way that complements, rather than competes with, other initiatives. In short, the Dialogue can act as a connector, accelerator, and legitimacy layer: connecting fragmented initiatives, accelerating practical convergence, and strengthening global trust in AI governance.

Different stakeholders can contribute most effectively if the Dialogue is structured to move from broad principles to practical implementation. Governments should bring policy priorities, regulatory realities, and lessons from national AI strategies. The private sector should contribute operational experience from deploying AI at scale, especially in regulated and high-impact environments. Academia and the technical community should provide evidence on safety, evaluation, interoperability, and lifecycle governance. Civil society should ensure that human rights, inclusion, accountability, and social impact remain central. To make that contribution meaningful, I would recommend a format that is highly practical and outcome-oriented. The Dialogue should combine plenary sessions with smaller expert working groups focused on specific thematic clusters, so that discussion can be translated into usable guidance rather than remaining at a high level. It should also include structured opportunities for written inputs, case studies, and pilot examples from different regions and sectors. That would help ground the Dialogue in real-world deployment challenges, not just abstract policy concepts. In terms of structure, the Dialogue should be inclusive but disciplined. A clear agenda, defined outputs, and a mechanism for follow-up are essential. It should aim to produce concrete deliverables such as shared terminology, interoperability principles, implementation insights, and capacity-building priorities. It would also be valuable to create continuity between meetings through a standing multi-stakeholder reference group or periodic progress updates. The most effective Dialogue will be one that listens broadly, but also narrows quickly toward actionable outcomes. If it is designed well, it can become a bridge between global policy ambition and practical AI governance on the ground.

Underrepresented voices include the Global South, local and Indigenous communities, women, young people, people with disabilities, and other groups most affected by AI but least represented in global policy forums. The biggest gap is that governance discussions are still too often shaped by large technology firms and institutions in the Global North, while the lived realities of communities dealing with data extraction, labour impacts, infrastructure dependence, language exclusion, and weak regulatory capacity are underweighted. They can be included by moving from consultation to co-design: funding participation, translating materials, supporting regional and community-led forums, and ensuring representation is not limited to capital cities or elite institutions. It also helps to create dedicated channels for civil society, marginalized communities, and technical experts from underrepresented regions to contribute on equal footing, rather than only through broad plenary sessions. A practical approach is to combine formal seats at the table with targeted capacity-building, travel support, and outreach through trusted local intermediaries. Another useful method is to gather input through real use cases from affected communities, so governance debates are grounded in how AI actually changes services, jobs, rights, and access to opportunity.

The most effective formats would combine structured discussion with practical, participatory methods that bring out real-world experience rather than only formal statements. A strong model would be to use short plenary sessions for shared framing, followed by smaller breakout groups focused on specific themes or use cases. This makes the Dialogue more dynamic and ensures that different stakeholder groups can contribute meaningfully. Another effective format would be case-based working sessions, where participants discuss concrete deployment scenarios from sectors such as healthcare, education, finance, or public administration. These sessions help move the conversation from abstract principles to operational governance questions. The Dialogue could also benefit from multi-stakeholder clinics or labs, where governments, technical experts, civil society and industry work together to solve a defined challenge, such as transparency, interoperability, or capacity-building. This would encourage practical cooperation and produce tangible outputs. In addition, regional roundtables or hybrid participation formats would improve inclusion, especially for stakeholders who may not be able to travel. Structured written inputs, pre-meeting consultations, and moderated online participation could help ensure broader geographic and sectoral representation. Finally, the Dialogue should include a mechanism for follow-up and iteration, so that each meeting builds on the last. The most valuable engagement formats are those that create continuity, accountability and visible progress over time.

Examples that are especially relevant are: Clear AI governance frameworks with named accountability, approval rights, risk tiering, and update processes. Lifecycle controls such as validation before launch, continuous monitoring in production, incident response, and decommissioning rules. Human-in-the-loop oversight for high-impact use cases, so AI supports rather than replaces accountable human decision-making. Transparency and explainability practices that document model purpose, limits, data sources, and decision logic in a way stakeholders can understand. Cross-functional governance committees bringing together business, legal, compliance, privacy, security, and technical teams. Responsible AI platforms that provide monitoring, bias detection, audit trails, compliance checks, and performance dashboards. Sector-specific controls for regulated environments, where governance is adapted to the risk level of healthcare, finance, public services, or critical infrastructure. A useful pattern is to combine policy with operational tooling: rules alone are not enough unless they are embedded into workflows, monitoring, and ownership. The strongest approaches make governance practical, repeatable, and measurable across the full AI lifecycle.