OS

Honestly, I think success starts with honesty; acknowledging that AI governance today is fragmented, unequal, and moving slower than the technology itself. The first outcome I would want to see is practical coherence. Not uniformity, but enough common ground that countries can work together without each having to build everything from scratch. Frameworks like ISO/IEC 42001 exist, but they need to be translated into something accessible for governments and organisations that do not have large technical teams. The second outcome is genuine inclusion. Not just representation in the room, but voices that actually shape the outputs. Communities whose languages, cultures, and livelihoods are most affected by AI systems are often the last to be consulted. That needs to change. Third, and perhaps most important — the Dialogue needs to produce something that can be followed up on. A conversation that ends with a communiqué and no mechanism for accountability is just a very expensive meeting. Success means leaving with clear next steps, named responsibilities, and a review process. If those three things happen, I think the Dialogue will have earned its place as the first of many ; rather than the last.

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

These four feel most urgent to me because they address both the foundation and the fault lines of AI governance. Safe, secure and trustworthy AI is where everything begins. If people - and institutions - cannot trust AI systems, nothing else works. I work with ISO/IEC 42001 and ISO/IEC 27001 standards, and even within those structured environments, building genuine trustworthiness takes sustained effort. It does not happen by declaration. Capacity-building matters because the governance conversation keeps assuming that everyone is starting from roughly the same place. They are not. There are countries and communities that are already being shaped by AI systems they had no hand in designing and no means of scrutinising. That gap has to be a priority, not a footnote. The social, cultural and linguistic implications of AI is the one I feel most personally about. I work at the intersection of technology and traditional heritage, and I have seen how easily AI systems overlook what does not fit neatly into a dataset. Language, craft knowledge, cultural identity - these things matter, and they are at risk if governance does not deliberately protect them. Transparency and human oversight complete the picture. Standards without accountability are just paperwork. Oversight means someone is actually responsible - and can be held to it.

One area I think deserves more attention is traditional and intangible cultural knowledge. AI systems are increasingly capable of replicating - and sometimes misrepresenting - craft techniques, oral traditions, indigenous practices, and heritage knowledge that communities have preserved across generations. This is not just a cultural sensitivity issue. It is a governance gap. These knowledge systems are not public domain, and current frameworks do not adequately protect them. The second thing I would raise is the situation of smaller actors - artisans, cultural practitioners, community-based organisations, small businesses. Most of the governance conversation assumes large institutions with legal teams and compliance functions. But the people most exposed to AI disruption are often the ones least equipped to navigate it. Governance frameworks need to reach further down. And finally, I think we need to talk more honestly about the human experience inside AI-mediated environments - not just the systemic risks, but what it actually feels like to trust, or distrust, or depend on systems you cannot fully understand. That dimension tends to get left out of technical governance discussions, but it may be where the most important questions are.

I am based in Malaysia, and I work across two worlds that do not often appear in the same conversation — AI governance and traditional cultural heritage. That dual perspective shapes how I see this question. The most immediate challenge I observe is the gap between policy ambition and practical implementation. Malaysia has made meaningful commitments to AI development, but governance frameworks have not kept pace. Organisations — particularly smaller ones — are adopting AI tools without structured guidance on risk, accountability, or ethical deployment. Awareness of standards like ISO/IEC 42001 remains low outside of large enterprises and regulated sectors. In the heritage and creative sector specifically, the governance gap is almost invisible — because no one is really looking there. AI tools are already being used to generate content that resembles traditional crafts, batik motifs, songket patterns, and other forms of intangible cultural heritage. There is no clear framework to address misappropriation, misrepresentation, or the erasure of the human mastery behind these traditions. The communities most affected are the least represented in governance conversations. For the broader Southeast Asian region, I think the most significant challenge is that most governance frameworks being discussed globally were designed with very different contexts in mind. They assume institutional capacity, legal infrastructure, and technical literacy that many countries in this region are still building. Adopting them wholesale creates compliance burden without creating genuine protection. The opportunity, though, is real. Malaysia and the region have a chance to contribute governance perspectives that are genuinely different — rooted in cultural plurality, community-based knowledge systems, and a development context that the Global North does not speak for. If the Dialogue creates space for that, it could produce frameworks that actually work for more of the world.

The most valuable thing the AI Dialogue can do is create a space where countries are not just presenting their positions, but actually learning from each other. Right now, international cooperation on AI governance tends to happen between countries that already have significant technical and institutional capacity. The conversations are important, but they leave out a large part of the world. The Dialogue, sitting within the UN system, has a legitimacy that other forums do not — and it should use that to bring genuinely different perspectives into the room. Practically, I think the Dialogue can play three roles. First, it can serve as a translation layer — helping countries understand how different governance frameworks relate to each other, where they are compatible, and where they genuinely conflict. This is not about forcing consensus. It is about making cooperation possible without demanding uniformity. Second, it can identify and support shared infrastructure. Things like mutual recognition of AI auditing standards, common principles for cross-border data governance, and coordinated approaches to high-risk AI applications — these require international agreement, and the Dialogue is well placed to initiate that work. Third, and perhaps most importantly, it can normalise the idea that governance is an ongoing practice, not a one-time declaration. AI is moving fast. What the Dialogue establishes today will need to be revisited. Building in review mechanisms and regular follow-up from the start would signal that this is a serious, long-term commitment.

There are several important initiatives already underway that the Dialogue should connect with rather than duplicate. UNESCO's Recommendation on the Ethics of AI is one of the most inclusive frameworks developed so far — it was adopted by all member states and covers cultural, social, and environmental dimensions that purely technical frameworks often miss. The Dialogue should treat it as a foundation rather than starting from scratch. The OECD AI Principles and the work of the Global Partnership on AI have produced useful guidance, particularly around transparency and accountability. But their reach is limited to countries with the capacity to engage fully with them. The Dialogue can add value by helping extend that work to contexts where implementation support is needed, not just principles. ISO/IEC 42001, the international standard for AI management systems, represents years of technical consensus-building. It gives organisations a structured way to govern AI responsibly. The Dialogue could acknowledge this standard and encourage its adoption as part of a broader governance ecosystem — particularly for smaller nations that benefit from ready-made frameworks. The added value the Dialogue brings is something none of these initiatives have on their own: universal membership, political weight, and the ability to connect governance to the broader UN agenda on sustainable development, human rights, and cultural preservation. That combination is rare. The Dialogue should use it deliberately — not to create yet another framework, but to give existing work the reach and legitimacy it needs to matter everywhere, not just in well-resourced capitals.

The Dialogue needs to be designed so that contribution is genuinely possible — not just symbolically open. Right now, most international governance forums are structured in ways that favour large delegations, well-funded organisations, and participants who are already embedded in global policy networks. If the AI Dialogue replicates that structure, it will get the same voices it always gets. My recommendation is a tiered participation model. Governments and intergovernmental bodies would naturally anchor the formal proceedings. But alongside that, there should be dedicated tracks for civil society, practitioners, researchers from the Global South, and community representatives — with real input mechanisms, not just observer status. What people say in side events should have a visible path into the main outcomes. For format, I would suggest moving away from panel-heavy plenary sessions toward smaller, problem-focused working groups. The questions in AI governance are technical enough that broad declarations rarely capture what actually needs to be resolved. Smaller groups with clear mandates tend to produce more honest and more useful outputs. There should also be a pre-Dialogue consultation process that is genuinely accessible — in multiple languages, available online, and structured so that individuals and small organisations can contribute meaningfully without needing a team of policy writers behind them. This form is a step in the right direction.

Several communities come to mind immediately, and I say this not as an abstract observation but from experience working across both the technology governance space and the traditional heritage sector in Southeast Asia. Traditional knowledge holders and artisan communities are almost entirely absent from AI governance discussions. These are people whose livelihoods, identities, and cultural knowledge are directly at risk from AI systems that replicate, commodify, or misrepresent what they have spent lifetimes preserving. They rarely have the language or the access to participate in UN forums — but they have some of the most grounded and urgent perspectives on what AI governance needs to protect. Rural and indigenous communities, particularly across Asia, Africa, and Latin America, are similarly missing. AI is not an abstract future for them — it is already affecting agricultural decisions, healthcare access, and economic opportunity. Their experiences should be informing governance, not arriving as case studies after the fact. Small and micro enterprises in the creative and cultural sectors are another overlooked group. They are navigating AI adoption without guidance, without protection, and without anyone really asking what they need. To include them, the Dialogue needs accessible consultation channels, translated materials, and partnerships with organisations already trusted by these communities. It also needs to be honest that inclusion takes more effort than opening a web form.

The most effective formats I have seen — and experienced — are the ones that combine structure with genuine conversation rather than replacing one with the other. A few ideas worth considering: Regional pre-dialogues, hosted in different parts of the world in local languages, could gather perspectives before the main event rather than after. This shifts the Dialogue from a reporting exercise to an actual input process. Practitioner showcases alongside the formal sessions could give space to people doing real work on the ground — not presenting papers, but sharing lived experience with AI's implications in their specific contexts. This kind of testimony tends to cut through in ways that policy language does not. Deliberative formats — where small mixed groups of participants work through a specific scenario or dilemma together — can surface genuine disagreement and nuance in ways that panel discussions rarely achieve. The goal is not consensus but clarity about where the real tensions lie. And for those who cannot attend in person, asynchronous contribution mechanisms — recorded input, written submissions with genuine turnaround, online working groups — would extend meaningful participation beyond whoever can afford to travel. The measure of an innovative format is not how creative it looks. It is whether the people who needed to be heard actually got heard.

A few examples stand out to me - some from international standard-setting, some from regional practice, and one from my own work. ISO/IEC 42001:2023 is probably the most practical governance tool available to organisations right now. It gives a structured, auditable framework for managing AI responsibly - covering risk, transparency, accountability, and continual improvement. What I appreciate about it is that it is not just a checklist. When implemented properly, it changes how an organisation thinks about AI, not just how it documents it. I hold a Lead Auditor certification in this standard, and I have seen firsthand how it creates a common language between technical teams, leadership, and auditors. Singapore's Model AI Governance Framework is another example worth noting. It is one of the more practical national-level frameworks produced in this region - written in a way that organisations can actually use, not just governments. It has been updated over time and has a testing sandbox component that allows responsible experimentation. That combination of guidance and space to learn is something more countries should adopt. The EU AI Act, for all its complexity, establishes something important: the principle that AI systems should be classified by risk level, and that higher-risk applications carry heavier obligations. That risk-tiering logic is sound and transferable even to contexts that would not adopt the Act itself. From my own context, I have been applying AI governance thinking within a heritage enterprise - assessing how AI tools interact with traditional craft knowledge, cultural authenticity, and community interests. It is a small-scale example, but it illustrates something the Dialogue should take seriously: governance is not only a large-institution concern. It needs to work at the community and enterprise level too, where most people actually encounter AI. The common thread in all of these is that effective governance combines clear principles with practical tools - and leaves room for context.