Breda University of Applied Sciences

Success would require moving beyond declaration toward actionable governance infrastructure. Three outcomes matter most. First, a shared definitional baseline. AI governance conversations are currently fragmented because stakeholders operate with incompatible understandings of what "AI readiness" actually requires. A meaningful outcome would be consensus language that distinguishes technical capacity from the organizational, cultural, and institutional conditions that determine whether AI deployment serves or harms communities in practice. Second, governance mechanisms that include under-resourced and professionally-oriented institutions. Universities of Applied Sciences across Europe serve students who will implement AI in society's critical functions, healthcare, engineering, education, logistics. These institutions are largely absent from global governance conversations. A successful Dialogue would establish participation pathways for applied and polytechnic higher education, not only research-intensive universities. Third, a commitment to interoperability over harmonization. The EU AI Act, UNESCO's Recommendation on the Ethics of AI, and emerging national frameworks approach governance from different philosophical traditions. Rather than forcing convergence, the Dialogue should produce mechanisms for mutual recognition and cross-framework translation, enabling institutions operating across jurisdictions to act coherently without waiting for global consensus that may never arrive. As an institution that has deployed AI across approximately 7,500 users and developed a governance framework recognized by UNESCO, we observe that the gap between policy intent and institutional reality is substantial. Governance that does not account for organizational and cultural dimensions of implementation will not close that gap

• AI capacity-building
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

Our four priorities reflect both the evidence from BUas's institution-wide AI deployment and broader patterns we observe across European higher education. AI capacity-building is foundational but currently too narrowly defined. Most frameworks address technical skills while the organizational conditions required for responsible adoption, governance structures, leadership capacity, institutional culture, and change management, remain largely unaddressed. Capacity-building that targets individuals without attending to institutional context produces uneven adoption, not transformation. Social, economic, ethical, cultural, linguistic and technical implications are inseparable in practice. Deployment across a professionally-oriented student population, who will work in regulated, human-centered sectors, shows that ethical implications are not abstract: they surface immediately in curriculum design, assessment integrity, and professional identity formation. Transparency, accountability, and human oversight are necessary but insufficient if framed solely as technical controls. At institutional level, these principles require governance structures, role clarity, and cultural conditions that most institutions have not yet developed. Accountability without organizational readiness produces liability, not safety. Interoperability of governance approaches is the priority most underserved by current frameworks. Institutions operating across the EU, navigating the AI Act, UNESCO recommendations, and national frameworks simultaneously, experience these as divergent rather than complementary. Practical interoperability mechanisms are urgently needed for institutions that cannot wait for global convergence.

The most significant gap is the absence of organizational and institutional readiness as a governance category distinct from individual AI literacy or technical capacity-building. Current frameworks assume that if individuals have AI knowledge and appropriate guardrails are in place, responsible use follows. Deployment experience at scale suggests otherwise. Institutions exhibit collective patterns of AI avoidance, over-reliance, and strategic misalignment that are not reducible to individual skill deficits or policy gaps. These are organizational phenomena requiring organizational responses: governance architecture, cultural change processes, and leadership that models critical engagement rather than uncritical enthusiasm or blanket resistance. Several emerging frameworks, including work recognized by UNESCO in the context of higher education, have begun to map these dimensions, but they remain outside mainstream governance discourse. A second emerging issue is the governance of agentic AI in educational and professional formation contexts. As AI systems move from tools to agents capable of planning and executing across complex tasks, a governance question emerges that frameworks have not yet addressed: what capacities must students and early-career professionals retain independently, and who is responsible for ensuring they do? This is simultaneously a question of human rights, educational accountability, and long-term societal resilience. Global governance that omits the institutional readiness and agentic dimensions will remain structurally incomplete, effective at the level of standards and rights, but disconnected from the conditions under which AI is actually adopted and used.

In European higher education, the four governance gaps identified are producing concrete and measurable effects. The absence of organizational readiness as a governance category means that institutions are being held accountable for AI outcomes they lack the structural capacity to produce. BUas became the first university of applied sciences in the EU to deploy a large language model institution-wide, across approximately 7,500 users. That process revealed that the most significant barriers were not technical or legal, but organizational: unclear role distribution, inconsistent leadership engagement, absence of shared frameworks for what responsible use looks like in practice, and significant variation in staff confidence and institutional culture. None of these are addressed by current governance frameworks. The interoperability gap creates a compounding burden for European institutions. Navigating the EU AI Act, national implementation guidelines, UNESCO recommendations, and institutional policies simultaneously, without translation mechanisms between them, consumes disproportionate capacity in institutions that are already under-resourced for AI governance work. The narrow definition of capacity-building means that funding and support flows toward technical training while the change management, leadership development, and governance architecture work goes unfunded and unrecognized. The opportunity lies precisely in the applied higher education sector. Universities of Applied Sciences educate the practitioners, nurses, engineers, teachers, logistics professionals, artists, who will determine how AI functions in society's everyday infrastructure. This sector has both the motivation and the proximity to translate governance principles into professional formation. With appropriate recognition and support, it can serve as a critical bridge between global governance frameworks and real-world implementation. That role is currently invisible in governance design.

The AI Dialogue can play a role that no existing mechanism currently fills: creating structured conditions for governance learning across jurisdictions, not just governance alignment. Most international AI cooperation efforts aim at convergence, shared standards, common definitions, aligned regulation. This is necessary but insufficient. What institutions, educators, and practitioners need is the ability to learn from each other's implementation experience across different regulatory and cultural contexts. A university in the Netherlands operating under the EU AI Act has lessons for a counterpart in Brazil or South Africa, not because the regulatory contexts are the same, but because the organizational and cultural challenges of responsible AI adoption are remarkably similar regardless of jurisdiction. The Dialogue can advance cooperation in three concrete ways. First, by establishing a permanent knowledge exchange infrastructure, not a one-time event but an ongoing mechanism through which institutions and sectors can share implementation evidence, governance instruments, and failure cases without competitive disadvantage. Second, by creating formal roles for civil society, education, and applied research institutions alongside governments and industry, ensuring that governance insights from the sector closest to AI's everyday social impact are systematically included. Third, by developing a governance translation layer: practical tools that help institutions map their local regulatory obligations onto shared principles, reducing duplication of effort and enabling genuine cross-border collaboration. International cooperation on AI governance will remain fragile if it operates only at the level of states and corporations. The Dialogue has an opportunity to build the infrastructure for a broader, more durable form of cooperation, one that includes the institutions that actually shape how AI is used in society.

Several existing initiatives provide foundations the Dialogue should actively build upon rather than duplicate. UNESCO's Recommendation on the Ethics of AI (2021) and its associated readiness assessment methodology offer the most comprehensive normative framework currently available, with genuine global reach. The Dialogue should treat this as a baseline, not a starting point for renegotiation. Similarly, the EU AI Act represents the most operationalized attempt to translate AI ethics principles into binding governance obligation, its implementation experience, including the tensions and practical difficulties it surfaces, is a resource for the global community. In higher education specifically, networks such as SURF in the Netherlands, the EU's Digital Education Hub, and emerging European university alliance AI working groups have developed practical governance instruments that have been tested at institutional scale. These are not visible in global governance conversations but represent exactly the kind of implementation knowledge the Dialogue needs. The added value the Dialogue can bring is threefold. First, it can function as a legitimizing bridge, connecting regional and sectoral governance work to a global framework that gives it political weight and cross-border recognition. Second, it can create accountability without uniformity: a mechanism through which jurisdictions and institutions report on governance progress against shared principles while retaining contextual flexibility. Third, it can explicitly elevate the role of education in AI governance, not merely as a sector to be governed, but as the primary site where the next generation of AI practitioners, policymakers, and citizens develop the capacities governance frameworks ultimately depend upon. Without that last function, AI governance will always be catching up to AI deployment.

The AI Dialogue's format will determine whether it produces genuine governance progress or another layer of declaration. Three structural recommendations follow from implementation experience in higher education. First, the Dialogue should be designed as a multi-year process with a permanent secretariat, not a single convening. AI governance challenges are not static, they evolve with deployment scale, model capability, and social context. A one-time event produces commitments; a sustained process produces learning. Each cycle should include a formal review of what governance instruments worked, what failed, and why. Second, stakeholder contribution needs to be structured by function, not just by sector. Governments bring regulatory authority; industry brings deployment scale and technical knowledge; civil society brings rights and equity perspectives; and education brings formation, the capacity to shape how future practitioners understand and exercise AI. These are distinct contributions requiring distinct participation formats. Roundtables that mix all stakeholders in undifferentiated discussion tend to amplify the most resourced voices. Structured working streams with dedicated rapporteurs for each function would produce more substantive input. Third, the Dialogue should establish a practitioner track alongside the policy track. Institutional implementers, university leaders, hospital administrators, public sector managers, have governance knowledge that is currently invisible at global level. They know what compliance costs in practice, where frameworks create perverse incentives, and what conditions actually enable responsible adoption. Mechanisms to systematically capture and integrate this knowledge, structured case submissions, regional practitioner panels, implementation evidence repositories, would significantly improve the quality of governance outputs. Format should follow function: the goal is durable governance infrastructure, not a successful event.

Global AI governance discussions are structurally biased toward three profiles: large technology companies, research-intensive universities, and governments from high-income countries with advanced regulatory capacity. This leaves several critical perspectives systematically underrepresented. Applied and vocational higher education institutions are largely absent. Yet universities of applied sciences, polytechnics, and community colleges educate the majority of students who will implement AI in practice, in healthcare, engineering, social work, education, tourism and public administration. Their governance challenges are different from research universities: they operate closer to professional formation, with tighter resource constraints and more direct accountability for graduates' workplace readiness. Their absence from governance conversations means that frameworks are designed without adequate consideration of how AI is actually adopted in applied professional contexts. Early-career practitioners and students are structurally excluded. These are the people who will live with AI governance decisions for decades. Mechanisms for structured and genuine youth and student input are consistently missing. Institutions from the Global South face a compounded disadvantage: they are under-resourced for governance participation and their implementation contexts are poorly represented in the frameworks being developed. Governance designed primarily from high-income, high-capacity contexts will not transfer cleanly to environments with different infrastructure, cultural norms, and regulatory traditions. Inclusion mechanisms that would make a concrete difference: dedicated funding for participation from under-resourced institutions; regional pre-dialogues that aggregate and synthesize input before global convenings; translation support that is substantive rather than merely linguistic; and formal requirements that governance outputs include implementation guidance for low-resource contexts. Representation without structural support for participation is symbolic, not substantive.

The most important design principle is that format should generate usable governance knowledge, not just visible participation. Several formats have demonstrated effectiveness in adjacent contexts and warrant consideration for the AI Dialogue. Structured peer exchange, similar to the Braindate model used in professional learning conferences, allows participants to self-organize around shared governance challenges rather than passively receiving presentations. When practitioners from different jurisdictions and sectors can identify counterparts facing equivalent problems, the quality of knowledge exchange significantly exceeds what panel formats produce. A governance dialogue that builds in dedicated time for this kind of structured peer connection will generate insights that no plenary session can. Deliberative working sessions with published outputs should replace or substantially reduce traditional panel formats. Panels produce visibility for speakers; deliberative sessions produce commitments and documented reasoning. If each thematic area concludes with a rapporteur-synthesized output that feeds directly into Dialogue conclusions, the connection between participation and governance outcomes becomes traceable. Asynchronous and multilingual engagement infrastructure is a prerequisite for genuine inclusivity, not an add-on. Pre-Dialogue input mechanisms, structured online consultations, regional synthesis workshops, translated submission pathways, allow institutions and communities that cannot fund in-person attendance to contribute substantively. The Dialogue's legitimacy depends on this. Finally, implementation showcases, short, structured presentations from institutions that have deployed AI governance frameworks at scale, provide the kind of grounded evidence that policy discussions typically lack. These are distinct from academic presentations: the focus is on what worked, what failed, and what conditions made the difference. Engagement formats are themselves a governance signal. How the Dialogue is structured communicates what kinds of knowledge and whose experience it values.

Three examples from higher education practice illustrate what effective AI governance looks like when it moves from principle to implementation. The first is institution-wide deployment with structured governance architecture. Breda University of Applied Sciences (BUas) in the Netherlands deployed a large language model across approximately 7,500 students and staff, the first such deployment at a university of applied sciences in the EU. The governance approach combined a mandatory AI training programme built on clearly defined institutional learning outcomes, a two-lane assessment framework distinguishing AI-supported from AI-independent tasks, explicit role distribution across academic, professional, and leadership staff, and a dedicated Centre for Teaching and Learning with AI mandate. The result was not a policy document but an operational governance system tested at scale. The second is the AI Strategy Compass: a framework developed to guide institutional AI adoption across six components: urgency awareness, ambition and strategy, pioneer team development, programmatic approach, communication, and cultural change. Recognized by UNESCO, it provides a structured instrument for institutions to evaluate their governance maturity and identify gaps, moving beyond compliance checklists toward developmental governance. What these examples share is a commitment to governance as an organizational process rather than a regulatory event, ongoing, adaptive, and accountable to implementation evidence rather than to declaration alone.