Council on Licensure Enforcement and Regulation

A successful first Global Dialogue would not be measured by declarations signed or communiqués issued. It would be measured by whether it changed what regulators actually do the week after they return home. Three outcomes would constitute genuine success. First, convergence on a shared vocabulary without enforced uniformity. The core failure risk in global AI governance is that jurisdictions talk past each other because they have not agreed on what words like risk, explainability, and oversight mean in operational terms. A successful Dialogue would produce a common lexicon, not a common rulebook. Alignment, not uniformity. Second, movement from AI literacy to AI fluency. The CLEAR Principles identify capacity as the central implementation challenge. But the conceptual work must go further: the global conversation needs to shift from asking whether regulators understand AI to whether they can deploy that understanding strategically. The Dialogue would succeed if it produced a roadmap for building regulatory institutions that can lead, not just respond. Third, a meaningful commitment to jurisdictional inclusion. Too many global governance conversations produce frameworks built on the realities of high-resource, high-capacity systems and then invite everyone else to comply. A successful Dialogue would centre the experience of regulators operating under resource constraints, and ensure the governance architecture that emerges is genuinely portable across legal traditions, infrastructure levels, and professional contexts. Underneath all three is a harder test: did the Dialogue produce trust? Not trust in AI. Trust between the institutions that govern it. Without that, no framework holds. The first Global Dialogue will be judged not on ambition but on whether it laid foundations that the second one could build on.

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

AI systems do not recognise borders. They do not pause at customs. They do not defer to national legislation or jurisdictional tradition. A model trained in one country, deployed in another, and affecting citizens in a third operates in a governance vacuum that no single regulator, and no single government, can fill alone. That is why these four priorities are not merely preferences. They are prerequisites. Safe, secure and trustworthy AI is the foundation. Without it, nothing built on top holds. But safety without shared definition is theatre. What one jurisdiction calls a safeguard, another calls a barrier to innovation. The Dialogue must close that gap. The social, economic, ethical, cultural and technical implications of AI remind us that this is not a technology conversation. It is a civilisational one. The communities most likely to be harmed by poorly governed AI are rarely the communities designing it. Any framework that does not centre that asymmetry will fail the people it claims to protect. Interoperability of governance approaches is where ambition meets reality. We did not ask every nation to adopt identical legal systems. We asked them to recognise each other's judgments, honour shared norms, and build mechanisms for cooperation when things go wrong. AI governance requires the same architecture. The International Atomic Energy Agency did not eliminate nuclear risk. It created the conditions under which risk could be collectively managed. We need that equivalent here, and we need it now. Transparency, accountability and human oversight are not technical requirements. They are democratic ones. When an algorithm shapes a decision, a citizen must be able to ask why. When it causes harm, someone must be answerable. The Geneva Conventions did not emerge from optimism. They emerged from the recognition that without common rules, catastrophe becomes inevitable. We are at that moment again.

You are not wrong. You are early. Here is the 300 words: The themes identified in Resolution 79/325 reflect a maturing conversation about governance. But governance, however well designed, is a preventive architecture. What is missing from the current framing is an equally serious conversation about response. Every governance framework assumes a degree of good faith from the actors it governs. It assumes systems will be built within recognisable institutional structures, subject to identifiable accountability chains, and amenable to regulatory intervention. Those assumptions do not hold for every scenario this Dialogue must eventually confront. Non-state actors. Autonomous systems operating across jurisdictions without a responsible sovereign. AI deployed not to serve populations but to destabilise them. Financial markets manipulated at a speed no human regulator can match. Electoral systems undermined by systems no election commission has the tools to audit. These are not hypothetical risks. They are emerging ones, and the current thematic architecture has no clear home for them. The International Atomic Energy Agency does not simply promote the peaceful use of nuclear technology. It provides a mechanism through which the international community can identify, escalate, and respond to threats that individual states cannot manage alone. We have no equivalent for AI. The United Nations has existing mechanisms for collective security response. What it lacks is the technical doctrine, the early warning infrastructure, and the agreed threshold for intervention that would allow those mechanisms to function effectively in an AI context. The cross-cutting issue that is missing is not governance. It is deterrence, detection, and coordinated international response to AI-enabled threats that cross the threshold from regulatory concern to security emergency. The first Global Dialogue will be judged, in time, not only on the quality of its governance frameworks. It will be judged on whether it had the foresight to ask what happens when those frameworks are insufficient. That conversation needs to begin here.

CLEAR membership has over 550 professional regulatory bodies across four continents. That breadth is precisely what makes governance gaps visible in ways that single-jurisdiction perspectives cannot capture. What our membership reveals is not a story of global progress at different speeds. It is a story of divergence that is accelerating. The European Union's AI Act represents the most comprehensive attempt to date to create a risk-based legislative framework with teeth. Australia has moved with similar seriousness, embedding AI governance within existing regulatory obligations while developing sector-specific guidance. The United Kingdom has pursued a principles-based approach that prioritises adaptability over prescription. Each of these reflects genuine institutional commitment and, critically, a willingness to accept that governance has a cost. The challenge is that AI does not operate within these regional architectures. It operates across them. A professional regulatory body in a jurisdiction with robust AI governance still relies on systems built, trained, and deployed elsewhere, often in environments where no equivalent framework exists. The asymmetry is not theoretical. It is operational and it is daily. For professional regulators specifically, the governance gap creates three compounding problems. First, they cannot always know whether the AI tools they oversee meet standards their own jurisdiction would require. Second, the professionals they license may be using systems whose provenance and risk classification are entirely opaque. Third, when things go wrong, the accountability chain dissolves at the jurisdictional boundary. The opportunity is equally significant. CLEAR's membership spans the full spectrum of regulatory maturity. That means it also spans the full spectrum of practical knowledge about what works, what fails, and what smaller or less-resourced regulators actually need to implement governance that is meaningful rather than merely documented. The gap between the most and least governed AI environments is where the greatest risk lives. It is also where international dialogue can make the greatest difference.

Let me be direct about what the AI Dialogue is, and what it is not. It is not a legislature. It cannot compel. It cannot enforce. It cannot substitute for the hard political work that happens inside national parliaments, regional bodies, and regulatory institutions. Anyone who arrives in this Dialogue expecting it to produce binding global law will leave disappointed, and they should manage that expectation now. But here is what it can do. And what it can do matters enormously. It can create the conditions under which cooperation becomes the rational choice. Not through obligation, but through architecture. Through shared vocabulary. Through mutual recognition. Through the slow, unglamorous work of building the trust that precedes every meaningful international agreement ever reached. The history of international cooperation is not a history of nations surrendering sovereignty to good ideas. It is a history of nations recognising that certain risks are too large, too fast, and too borderless to manage alone. Nuclear technology. Climate. Pandemic preparedness. In each case, the breakthrough was not a single treaty. It was the establishment of a common frame of reference from which treaties eventually became possible. AI governance needs that common frame. The Dialogue can build it. It can surface what 550 professional regulatory bodies across four continents have already shown to be true: that the gap between the most and least governed AI environments is not a gap that market forces will close. It requires deliberate, sustained, internationally coordinated effort. It can also do something less visible but no less important. It can give smaller jurisdictions, under-resourced regulators, and communities furthest from the centres of AI development a seat at the table before the architecture is fixed, not after. The Dialogue will not solve this. But it can make solving it possible. That is not a small thing. That is everything.

For over a decade, the international community has been building AI governance from the middle outward. The OECD AI Principles established a normative reference that over fifty countries have endorsed. The WHO Expert Working Group on Regulatory Considerations of AI for Health has developed a Template Law and training architecture to support Member States in building national regulatory frameworks aligned in principle while sovereign in design. The European Union's AI Act has translated principles into binding legislative obligations with genuine enforcement architecture. The United Kingdom has pursued adaptive, regulator-led frameworks that prioritise institutional judgment over statutory prescription. CLEAR's Principles for Ethical and Effective AI in Professional Regulation have extended that work to professional oversight bodies across four continents. Each of these represents serious, consequential work. None of it is wasted. But all of it has been built in the absence of the one thing that should have come first: an overarching international framework from which national, regional, and sectoral approaches could derive coherence, rather than seeking it retrospectively. We have been working in reverse. And the consequence is visible in the divergence between jurisdictions, the inconsistency between sectoral frameworks, and the governance vacuum that exists precisely where AI systems operate most freely, which is across the boundaries none of our existing architectures were designed to govern. The AI Dialogue's most significant contribution would be to correct that inversion. Not to replace what has been built, but to provide the overarching architecture from which existing frameworks can draw legitimacy, alignment, and mutual recognition. The OECD, WHO, EU, UK, and CLEAR have each built something real. The Dialogue can provide what they were each forced to build without: the common ceiling from which everything else can properly hang. That is not duplication. That is foundation work. And it is overdue.

Honesty requires acknowledging what multi-stakeholder dialogue tends to produce when it is not deliberately designed to avoid its own failure modes. It produces representation without influence. Civil society in the room but not in the draft. Industry at the table with better resourced delegations than most Member States. Academic voices heard once and not heard again. And smaller jurisdictions, those with the most to gain from genuine international alignment, present but peripheral. If the AI Dialogue reproduces that architecture, it will generate a communiqué. It will not generate change. Here is what would make it different. First, separate the conversation from the negotiation. The Dialogue needs genuine intellectual space, sessions where participants are not performing positions but actually thinking together. That requires design, not just agenda-setting. Second, weight participation by exposure to consequence, not by resource. The jurisdictions and sectors most affected by AI governance gaps are rarely those with the largest delegations. Structural mechanisms must ensure their experience shapes outcomes, not merely decorates them. Third, bring operational knowledge into the room alongside policy authority. Regulators who are making AI governance decisions today, not designing frameworks for tomorrow, have evidence that no think tank paper captures. CLEAR's membership alone has over 550 bodies doing this work in real time across four continents. That knowledge needs a formal channel into the Dialogue, not a side event. Fourth, be explicit about what the Dialogue will not resolve. Credibility requires honesty about scope. A Dialogue that overclaims will be dismissed. One that is precise about what it can deliver, and delivers it, builds the trust that makes the next conversation possible. The format that serves this Dialogue best is not the one that looks most inclusive on paper. It is the one disciplined enough to actually produce something.

The voices most underrepresented in global AI governance discussions are not hard to identify. They are hard to prioritise. And that distinction matters, because the problem is not ignorance. It is architecture. Professional regulators are largely absent. Not governments. Not ministries. The people who make daily decisions about whether a licensed professional can use an AI tool, whether an AI-assisted outcome meets the standard of care, whether a complaint arising from algorithmic error falls within existing jurisdiction. These are the practitioners of governance, and they are almost entirely invisible in global dialogue. Yet they are the implementation layer on which every framework ultimately depends. Many lower-income and developing nations participate in these conversations without the delegations, technical staff, or advance preparation time that meaningful engagement requires. Presence is not the same as influence. When governance frameworks are substantially shaped before they reach consultative stages, the window for genuine contribution has already closed. Structural reform of how and when these countries are brought into the process is not a courtesy. It is a prerequisite for frameworks that will actually function across the full range of contexts they claim to govern. Smaller and under-resourced regulatory bodies face a compounding challenge. They are asked to implement governance they did not design, for populations they know better than anyone who drafted the relevant framework. Their operational knowledge is precisely what global dialogue most needs and most consistently fails to capture. Indigenous communities, linguistic minorities, and populations underrepresented in AI training data have a direct and material stake in governance outcomes. Their inclusion requires more than invitation. It requires resource, translation, and structural guarantees that contribution shapes outcome rather than simply appearing in the record. The Dialogue should hold itself to one honest test. Are the people most exposed to the consequences of failure genuinely shaping the response?

Structured adversarial sessions. Not debate for performance, but deliberate challenge. Take a proposed governance principle and assign serious, informed participants to break it. Find the jurisdiction where it fails. Find the actor it does not reach. Find the scenario it cannot handle. Principles stress-tested in public carry more credibility than principles endorsed by consensus in a room where nobody wanted to be the dissenting voice. Regulatory practitioner tracks running parallel to the policy tracks. Not as side events. As equal constituents of the main programme. The people implementing governance in real time, across diverse jurisdictions and resource levels, have operational evidence that no policy paper contains. That evidence should inform the Dialogue's outputs, not its fringe programme. Red team exercises on governance gaps. Convene small, mixed groups, regulators, technologists, civil society, industry, and ask them to identify specifically how the proposed frameworks fail. Where are the loopholes. Where does accountability dissolve at a jurisdictional boundary. Where does a non-state actor simply walk through the architecture unchallenged. The findings should be published, not sanitised. And one structural commitment above all others: every session that produces a recommendation should also produce a named accountability mechanism for following it up.

The evidence base for effective AI governance already exists. The problem is not a shortage of examples. It is a failure to connect them into a coherent international learning architecture. The WHO Expert Working Group on Regulatory Considerations of AI for Health has developed a Template Law and accompanying training curriculum designed to support Member States in building national regulatory frameworks that are sovereign in design but aligned in principle. Critically, it is built around the reality of variable institutional capacity, not the assumption of well-resourced regulatory bodies. It is one of the few international instruments that takes implementation seriously as a design constraint rather than an afterthought. The OECD AI Principles and accompanying Policy Observatory provide the most comprehensive cross-jurisdictional mapping of AI governance approaches currently available. The Observatory does not prescribe. It illuminates. It allows policymakers and regulators to see how others have approached equivalent problems and to draw lessons without importing unsuitable solutions. The European Union's AI Act demonstrates that risk-based legislative classification is achievable at scale. Its most instructive contribution is not the framework itself but the process of building it: the consultations, the impact assessments, the iterative engagement with sector-specific regulators that produced a framework with genuine enforcement architecture rather than aspirational language. The CLEAR Principles for Ethical and Effective AI in Professional Regulation offer something the larger frameworks cannot: a governance model designed for the full spectrum of regulatory contexts, tested against the operational realities of over 550 professional regulatory bodies across four continents. They demonstrate that values-driven guidance can be both internationally coherent and locally adaptable. What connects these examples is not their scale. It is their orientation. Each one was designed with implementation in mind, not just endorsement.