UNDP

The Dialogue sits within a landscape already crowded with principles, frameworks, and high-level commitments on AI. Its value is not in adding to that body of text. It is in making the gap between what frameworks say and what institutions can do the starting point. That gap is the Dialogue's real subject. From a governance and a development perspective, in much of the world, we see that AI systems are being deployed in public services — benefits allocation, border control, health diagnostics, law enforcement — by governments that may not yet have dedicated AI regulation, independent oversight bodies, or structured processes for asking what rights are at stake. The existing framework does not always reach them in an operational sense. The Dialogue should make this visible and take it seriously. A successful Dialogue would establish that AI governance is an institutional challenge as well as a technical one. The question is not only what rules should govern AI, but whether the institutions responsible for applying those rules — courts, regulators, procurement bodies, civil society — have the capacity and independence to do so. That reframing has consequences for everything the Dialogue discusses: capacity-building, human rights accountability, and what meaningful international cooperation looks like in practice. UNDP's work in the Europe and Central Asia region, including the Human Rights Impact of AI (HRIA) Toolkit, a practical methodology for assessing the human rights impact of AI systems, reflects this conviction: governance holds only when it is tested against real institutional conditions.

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

UNDP's Regional Programme for Europe and Central Asia works across the region on our development mandate, including through governance, rule of law, and human rights programming. In December 2025, we published the Human Rights Impact of AI Assessment (HRIA) Toolkit, a practical methodology for assessing how AI systems affect human rights, developed for use by public authorities, civil society, development practitioners and beyond. The Toolkit emerged from a consistent observation across the region: regulatory frameworks are at very different stages of development, and most institutions deploying AI in public services have no structured process for identifying who might be harmed. Our HRIA methodology directly addresses this gap - specifically the question of accountability: who is responsible when an AI-assisted decision harms someone, and what recourse exists. An AI system that cannot be audited cannot be assessed for rights impact, which is why transparency and accountability are inseparable from the human rights priority in our work. The social, economic, and ethical implications priority reflects the same regional reality. AI governance frameworks shaped primarily by technically advanced contexts do not always fit countries where biometric systems are deployed under weak data protection regimes, or where LLMs are used in public services without linguistic or cultural adaptation. We are actively working to bring the HRIA methodology into broader international circulation, including through engagement with government (data protection) authorities, academia, legal experts, the technology sector, and civil society.

One issue that cuts across all themes but is not named directly is the question of institutional preconditions for governance. Most frameworks assume that the institutions responsible for AI oversight - regulators, courts, procurement bodies - exist, are independent, and have the capacity to act. In a significant part of the world, and in much of the region where UNDP operates, these preconditions are partial or absent. Where judicial independence is limited, data protection authorities are under-resourced or constrained, and civil society faces legal or operational restrictions, AI governance frameworks may not function as intended. They can exist on paper without the institutional capacity to enforce them. This question is not just about skills, but about institutional independence and operating space. A related issue is the relationship between AI deployment and civic space. In parts of Europe and Central Asia, AI-enabled surveillance and content filtering are being used in ways that affect the ability of journalists and civil society to operate. This has direct implications for the multi-stakeholder participation that effective AI governance depends on.

In the Europe and Central Asia region, AI deployment in public services is advancing faster than the governance structures meant to oversee it. Governments across the region are rolling out AI-assisted tools for e-government services, benefits administration, and law enforcement. In many cases, there is no legal framework specifically governing these deployments, no independent body with a mandate to audit them, and no structured process for assessing their human rights implications before or during use. The EU AI Act is the primary reference point for AI governance in the region, but its reach is uneven. Countries with EU accession prospects are beginning to align with it. Countries without that prospect, or where rule of law conditions diverge significantly from EU norms, face the same technologies and the same governance challenges, with fewer institutional resources and less external pressure to address them. The opportunity is to build governance capacity alongside design and deployment rather than after harm has occurred. UNDP's HRIA Toolkit was developed in direct response to this gap: a structured methodology that governments and oversight bodies can apply at any stage, without waiting for comprehensive national legislation.

The Dialogue sits within a broader ecosystem of AI governance initiatives, each with a distinct mandate and constituency. From UNDP's perspective, working at the intersection of governance and development across diverse institutional contexts, the most valuable function the Dialogue can serve is to connect the actors developing norms with the actors doing implementation in lower-capacity settings. In our experience in Europe and Central Asia, the gap between an international commitment on AI governance and what a national institution can actually do with it is significant. The Dialogue, given its UN mandate and inclusive membership, is well positioned to surface that gap and treat implementation experience as substantive input rather than background context.

The landscape already has significant architecture. The UNESCO Recommendation on the Ethics of AI provides a shared ethical vocabulary with genuine global reach. The Council of Europe Framework Convention is the first binding international instrument and introduces rights-based accountability obligations. The ITU AI for Good platform has built a practitioner community. A gap that the Dialogue could help address is the connection between standard-setting and implementation in under-resourced governance contexts. The existing instruments were largely designed by and for institutional environments with functioning regulatory bodies, independent courts, and resourced civil society. Organisations working where those conditions are absent — including UNDP across our programme countries — work to adapt and apply frameworks to conditions that were not always anticipated in their design. The Dialogue's added value could be precisely to bring implementation experience from diverse governance contexts into the centre of the conversation alongside the norms discussion.

The thematic breakout format in the current structure is well suited to substantive exchange. For international organisations like UNDP, the most useful contribution is programming experience: what happens when governance frameworks meet real institutional conditions. That is a different input from legal analysis or policy advocacy, and the Dialogue's thematic discussions can create space for it alongside the expert and governmental interventions.

AI governance discussions are largely shaped by those who develop or regulate AI systems. The people most directly affected by AI decisions — particularly in public service contexts — are almost entirely absent. Individuals whose benefit applications are denied by automated systems, whose movement is tracked by AI-enabled surveillance, or whose access to services is mediated by algorithmic tools, have no structured channel into any international governance process. Their experience is the most relevant evidence for evaluating whether governance is working. Practically, creating structured channels for affected communities and constrained civil society, through regional consultations and civil society support mechanisms, would strengthen the inclusivity and quality of the Dialogue's outputs.

From our experience running capacity-building and peer-learning programmes, case-based discussion tends to produce more actionable insights than thematic debate. When participants work through a concrete governance scenario — rather than presenting positions on a theme — the exchange surfaces what actually happened, what failed, and why. This is harder to achieve with prepared statements. We recognise the organisers have deep expertise in convening design. This is offered as one practitioner perspective on what has worked in analogous contexts.

UNDP's Istanbul Regional Hub developed the Human Rights Impact of AI Assessment (HRIA) Toolkit, published in December 2025, as a direct response to the AI governance gap that our programming across the Europe and Central Asia region made visible. The Toolkit provides a five-step methodology covering planning and scoping, impact identification, risk analysis and prioritisation, risk management and mitigation, and monitoring and reporting. It is grounded in international human rights law, designed to be applied by non-technical users, and built to function across different legal and institutional contexts rather than assuming the existence of specific regulatory infrastructure. The Toolkit and an interactive platform supporting its application are publicly available. Capacity-building programmes and engagement with regional governance bodies are ongoing. The methodology is replicable and is actively being brought into dialogue with international standard-setting processes. The Toolkit and interactive platform are publicly available at: https://www.undp.org/eurasia/publications/human-rights-impact-ai-assessment-toolkit and https://hria.eu. We also shared some of the background thinking in a recent blog post: https://www.undp.org/eurasia/blog/when-ai-governs-who-guards-our-rights.