Bangladesh University of Professional

Success would not be measured by declarations alone — but by whether the dialogue plants the structural roots for accountable, inclusive, and enforceable AI governance at a global scale. First, the dialogue must produce a shared definitional baseline. Nations currently operate with fragmented understandings of terms like "high-risk AI," "algorithmic accountability," and "AI safety." Without semantic alignment, even well-intentioned frameworks talk past one another. A living multilateral glossary — maintained through a permanent secretariat — would be a tangible first deliverable. Second, meaningful success requires that the Global South and small island developing states are not merely present, but genuinely influential. AI's most consequential deployments — in surveillance, healthcare triage, and financial inclusion — are already reshaping lives in the developing world, yet those voices remain underrepresented in standard-setting bodies. Structural equity in the dialogue's architecture is non-negotiable. Third, the dialogue should establish a standing mechanism for cross-border incident reporting and AI risk intelligence sharing — analogous to what FIRST and ISACs have achieved in cybersecurity. AI harms rarely respect borders; governance cannot either. Fourth, private sector accountability must move beyond voluntary pledges. A credible outcome would include consensus on minimum transparency obligations — particularly around training data provenance, model auditing, and algorithmic impact assessments — that governments can transpose into domestic law. Finally, success means institutionalizing continuity. A one-time summit produces communiqués; a recurring, treaty-anchored forum produces norms. The first dialogue succeeds if it commits to a second — with measurable benchmarks reviewed between sessions. In short: shared language, equitable voice, operational cooperation, enforceable standards, and institutional permanence. Anything less is a conversation. These five pillars together would be a foundation.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights

These three themes are not merely adjacent priorities - they form an interdependent framework that reflects the full lifecycle of responsible AI deployment, from design to societal impact. Safe, secure, and trustworthy AI is the natural foundation of my professional work. With nearly two decades in cybersecurity - including roles at national CIRT level, with the United Nations, and on World Bank-funded digital governance projects - I have witnessed firsthand how AI systems introduced without adequate security controls become vectors for exploitation. Trustworthiness is not a feature to be added after deployment; it must be engineered from inception. This theme anchors every other ambition in AI governance. The social, economic, ethical, cultural, linguistic, and technical implications of AI resonate deeply because AI governance cannot be reduced to technical standards alone. Having worked across diverse national contexts, I have seen how AI systems trained predominantly on high-resource languages and Western cultural datasets systematically disadvantage populations in the Global South. The economic displacement risks, the erosion of cultural identity through homogenized AI outputs, and the ethical blind spots embedded in opaque algorithmic systems are all governance failures - not just technical ones. This theme demands that policymakers confront AI's full human footprint. Protection and promotion of human rights is the ultimate accountability test for any governance framework. AI systems are increasingly influencing decisions about employment, healthcare, criminal justice, and political participation. Without explicit human rights anchoring - including the right to explanation, the right to contest automated decisions, and protections against discriminatory profiling - AI governance risks legitimizing the very harms it claims to prevent. Together, these themes trace a coherent arc: build systems that are secure by design, assess their full societal impact, and ensure they remain accountable to fundamental human dignity.

Cross-Cutting and Emerging Issues Beyond the Listed Themes The listed themes provide a strong normative foundation, yet several cross-cutting issues risk falling through the gaps between them - issues that are already shaping AI's real-world trajectory. AI and conflict is perhaps the most urgent omission. Autonomous weapons systems, AI-enabled disinformation at scale, and the use of predictive analytics in military targeting are accelerating faster than any governance conversation has acknowledged. The intersection of AI with international humanitarian law demands dedicated attention - not as a subcategory of human rights, but as a distinct domain with its own red lines and verification challenges. Concentration of AI power is a structural governance risk that no single theme fully captures. A handful of private actors - predominantly from two jurisdictions - now control the foundational models, the compute infrastructure, and the data pipelines that underpin global AI development. This concentration threatens not only market competition but sovereignty, democratic pluralism, and the very inclusivity that governance frameworks aspire to protect. Antitrust and geopolitical frameworks are ill-equipped to address it alone. AI and environmental sustainability remains conspicuously absent. The energy and water demands of large-scale AI training and inference are significant and growing. Any governance framework that ignores AI's ecological footprint is incomplete - particularly when climate-vulnerable nations bear the environmental costs of infrastructure they do not control. The AI talent and governance capacity gap cuts across every theme. Most nations - especially in the developing world - lack the trained personnel to audit AI systems, draft technically grounded regulation, or participate meaningfully in international standard-setting. Governance frameworks without parallel investment in human capacity will remain aspirational documents rather than operational realities. These gaps are not peripheral - they are the terrain where the next generation of AI harms will emerge.

Governance Gaps and Their Impact on My Country, Region, and Sector Bangladesh and the broader South Asian region sit at a critical inflection point — rapidly adopting AI-driven systems in public service delivery, financial inclusion, and healthcare, yet operating largely without binding AI governance frameworks, dedicated regulatory bodies, or the technical capacity to audit what is being deployed. In the cybersecurity sector — where I have worked for nearly two decades — the governance gap is acutely visible. AI-powered threat actors are exploiting the asymmetry between sophisticated attack capabilities and underfunded national cyber defenses. Bangladesh has experienced significant cyber incidents targeting financial infrastructure and government systems. As adversaries increasingly weaponize generative AI for phishing, deepfake fraud, and automated vulnerability exploitation, the absence of AI-specific incident response protocols compounds existing security weaknesses. The social and linguistic implications are equally pressing. Bangladesh's primary language, Bangla, remains severely underrepresented in large language models and AI training datasets. This creates a compounding disadvantage — AI tools deployed in education, healthcare, and public administration default to English-language logic, embedding cultural and linguistic bias into critical systems serving over 170 million people. On human rights, AI-enabled surveillance tools are being adopted across South Asia with minimal transparency or legal safeguards. The risk of discriminatory profiling — particularly affecting marginalized communities — is real and largely unaddressed by existing legal frameworks. The opportunity, however, is significant. Bangladesh's young, digitally active population and growing technology sector represent genuine capacity for leapfrogging outdated governance models — if international dialogue produces frameworks that are genuinely adaptable to lower-resource contexts, not merely exported from high-income jurisdictions. The central challenge is speed: AI deployment is outpacing governance by years, and the gap is widening.

The AI Dialogue arrives at a moment when the international governance architecture is fragmented, under-resourced, and unevenly representative. Its most consequential contribution would be to function not as another talking forum, but as a genuine coordination infrastructure — one that connects existing initiatives, fills institutional gaps, and converts political commitment into operational cooperation. First, the Dialogue can serve as a multilateral bridge. The current landscape includes the OECD AI Principles, the EU AI Act, the US Executive Order on AI, the G7 Hiroshima Process, and UNESCO's Recommendation on AI Ethics — all operating in parallel with limited interoperability. The Dialogue is uniquely positioned to map these frameworks against one another, identify conflicts and convergences, and build toward a common reference architecture that smaller nations can adopt without reinventing the wheel. Second, it can institutionalize South-South cooperation on AI governance. Developing nations share common challenges — limited compute access, underrepresented languages, weak regulatory capacity — yet rarely coordinate systematically. The Dialogue could establish dedicated working groups pairing nations at similar development stages, enabling peer learning and joint standard-setting outside the shadow of major power competition. Third, the Dialogue can anchor a global AI incident notification regime. Just as the International Telecommunication Union coordinates spectrum and cybersecurity norms, the Dialogue could sponsor a voluntary-to-binding escalation pathway for reporting cross-border AI failures — algorithmic discrimination, autonomous system malfunctions, AI-enabled information operations — creating shared situational awareness that no single nation can achieve alone. Finally, it can legitimize civil society and technical community participation in governance processes historically dominated by states and large corporations. Meaningful cooperation requires diverse epistemic inputs — researchers, affected communities, and domain experts — not just diplomatic delegations. The Dialogue's value is proportional to its institutional ambition.

The AI governance landscape is not a blank slate. Numerous initiatives have generated valuable norms, methodologies, and institutional relationships. The Dialogue's added value lies not in duplication but in integration, elevation, and operationalization. Foundational frameworks to build upon include UNESCO's Recommendation on the Ethics of AI — the only globally adopted intergovernmental instrument on AI ethics — and the OECD AI Principles, which have been endorsed by over 40 countries and provide a policy-ready baseline. The G7 Hiroshima AI Process produced voluntary codes of conduct for frontier AI developers that, while non-binding, established precedent for private sector accountability. The Global Partnership on AI (GPAI) has generated substantive technical work on responsible AI that remains underutilized in formal governance settings. Regional mechanisms deserve explicit connection. The EU AI Act represents the most comprehensive binding framework to date and will exert significant extraterritorial influence. The African Union's Continental AI Strategy and ASEAN's AI Governance Framework reflect growing non-Western governance ambition. The Dialogue should treat these not as subordinate instruments but as co-equal inputs into a genuinely pluralistic global architecture. Sectoral and technical bodies — including the International Telecommunication Union's AI for Good platform, ISO/IEC's AI standards committee (JTC 1/SC 42), and NIST's AI Risk Management Framework — provide technical credibility the Dialogue itself cannot independently generate. Connecting political will to technical standards infrastructure is precisely where multilateral dialogue adds irreplaceable value. The Dialogue's distinctive added value is threefold: it can convene actors that existing initiatives cannot — particularly from the Global South and civil society; it can create accountability linkages between voluntary commitments and verifiable outcomes; and it can establish the political legitimacy needed to transition soft norms into binding international instruments. Coordination without duplication is the defining discipline.

Stakeholder Contributions and Recommended Structure for the AI Dialogue Effective AI governance cannot be achieved through state actors alone. The complexity, pace, and distributional impact of AI demand a genuinely multi-stakeholder architecture — one where different actors contribute distinct, irreplaceable value rather than occupying ceremonial roles. Governments bring regulatory authority, diplomatic legitimacy, and the capacity to translate norms into binding law. Their primary contribution should be political commitment backed by concrete national implementation plans — not declarations without accountability mechanisms. The private sector — particularly frontier AI developers — must contribute mandatory transparency disclosures: model cards, training data provenance, third-party audit results, and incident reports. Voluntary pledges have demonstrated limited effectiveness; structural participation requirements are necessary. Civil society organizations and affected communities provide ground-truth evidence of AI's real-world harms and benefits. Their participation must be resourced, not merely invited. Travel grants, translation support, and dedicated plenary time — not side-event marginalization — are the minimum conditions for meaningful inclusion. Academic and technical experts should anchor working groups on specific governance challenges — auditing methodologies, standards interoperability, risk classification — ensuring that political outcomes are technically defensible. International and regional organizations can serve as neutral conveners, secretariat functions, and implementation monitors — roles that no single government or corporation can credibly occupy. On format and structure, the Dialogue should adopt a three-tier architecture. A high-level plenary — meeting biennially — sets political direction and adopts outcome documents. Thematic working groups — meeting intersessionally — produce technical deliverables on specific issues such as auditing standards, compute governance, and incident reporting. A permanent secretariat maintains institutional memory, tracks implementation, and ensures continuity between sessions. Critically, working group outputs should feed directly into plenary decisions — preventing the common failure mode where technical expertise and political authority operate in parallel without ever genuinely connecting.

Underrepresented Voices in AI Governance and Pathways to Inclusion Global AI governance discussions remain disproportionately shaped by a narrow set of actors — predominantly from high-income, English-speaking nations, large technology corporations, and established academic institutions. The voices most affected by AI's consequences are frequently the least represented in the rooms where governance decisions are made. The Global South is the most significant structural gap. Africa, South Asia, Southeast Asia, and Latin America collectively represent the majority of the world's population and are increasingly subject to AI-driven systems in healthcare, finance, law enforcement, and public administration — yet contribute marginally to international standard-setting bodies. Inclusion requires funded participation, not merely open invitations. Dedicated delegation support, regional preparatory processes, and guaranteed speaking roles in plenary sessions are practical minimum requirements. Indigenous communities face distinct AI governance challenges — including the extraction of traditional knowledge into training datasets without consent, and the erosion of cultural identity through algorithmically homogenized content. Their inclusion demands recognition of collective data rights and governance frameworks that respect indigenous sovereignty alongside individual rights. Persons with disabilities are both disproportionate beneficiaries and victims of AI systems — gaining from accessibility tools while facing systematic exclusion from biometric systems, automated hiring platforms, and facial recognition. Disability-led organizations must be structural participants, not afterthoughts. Frontline workers displaced or monitored by algorithmic management systems possess experiential knowledge that no technical expert can substitute. Labor organizations should have formal standing in governance processes affecting employment. Young people and future generations bear the longest exposure to AI governance decisions made today, yet youth representation in formal dialogues remains tokenistic. Inclusion mechanisms must go beyond representation to genuine influence — co-authorship of outcome documents, veto rights over directly affecting provisions, and independent youth and civil society tracks with direct plenary linkage. Presence without power is not participation.

Traditional conference formats — keynote addresses, panel discussions, and negotiated communiqués — are poorly suited to the complexity and urgency of AI governance. The Dialogue should deliberately experiment with formats that generate genuine understanding, surface hidden disagreements, and produce actionable outputs rather than polished consensus language that obscures real divergence. Red team sessions — borrowed from cybersecurity practice — would assign participants the explicit task of identifying weaknesses, loopholes, and unintended consequences in proposed governance frameworks. Structured adversarial scrutiny produces more robust outcomes than consensus-seeking deliberation alone. Governments, civil society, and technical experts would each conduct independent red team exercises, with findings synthesized in plenary. Living policy simulations would place delegations inside realistic AI governance scenarios — a cross-border algorithmic discrimination incident, an autonomous system failure affecting critical infrastructure, a disputed AI-generated election narrative — requiring real-time coordination across jurisdictions. Simulations expose governance gaps that abstract discussion conceals and build the relational trust that formal negotiations require. Citizen deliberative panels — modeled on successful climate and constitutional assemblies — would bring randomly selected members of the public from diverse national contexts into structured dialogue with technical experts. Their outputs, reflecting lay perspectives on AI's social contract, would formally inform working group deliberations rather than being treated as consultative decoration. Asynchronous multilingual participation platforms would enable voices unable to attend in person — from remote communities, under-resourced institutions, or linguistically marginalized populations — to contribute substantively between sessions. AI-assisted translation, carefully human-reviewed, could dramatically expand the linguistic accessibility of participation without sacrificing nuance. Outcome accountability reviews — public sessions where governments report against previous commitments before new ones are adopted — would introduce a discipline of follow-through currently absent from most international AI forums. The format is not administrative detail. It determines whose knowledge counts.

Effective Policies, Practices, and Approaches in AI Governance Several concrete examples across regulatory, technical, and institutional domains demonstrate that effective AI governance is achievable - and offer replicable models for the Dialogue to amplify and adapt. The EU AI Act represents the most comprehensive binding regulatory framework to date. Its risk-tiered classification system - prohibiting certain applications outright while imposing graduated obligations on high-risk uses - provides a replicable legislative architecture that other jurisdictions can adapt to their constitutional and developmental contexts without wholesale adoption. NIST's AI Risk Management Framework offers a voluntary but technically rigorous methodology for organizations to identify, assess, and mitigate AI risks across the full system lifecycle. Its profile-based approach allows sector-specific customization - making it practical for healthcare, financial services, and critical infrastructure operators simultaneously. Singapore's Model AI Governance Framework demonstrates how a smaller nation can exercise significant normative influence through practical, business-friendly guidance. Its emphasis on explainability, human oversight, and accountability has been adopted and adapted across Southeast Asia, proving that governance leadership is not reserved for large economies. The algorithmic impact assessment - pioneered in Canada's Directive on Automated Decision-Making - provides a structured pre-deployment evaluation mechanism requiring government agencies to assess bias, explainability, and appeal rights before deploying automated systems affecting citizens. This model is directly transferable to international procurement standards. Sector-specific ISAC models from cybersecurity - Information Sharing and Analysis Centers - demonstrate how cross-sector, cross-border threat intelligence sharing can be institutionalized without compromising commercial sensitivity. An AI-specific incident sharing architecture modeled on this approach would address the current absence of systematic cross-border AI harm reporting. Bangladesh's national cybersecurity incident response infrastructure, developed with international partners, illustrates how capacity-building investment enables developing nations to participate meaningfully in global governance rather than merely receive its outputs. Proven models exist. The governance gap is not imagination - it is political will and institutional commitment.