Bangladesh Internet Governance Forum

In the Bangladesh context, the first Global Dialogue on AI Governance would be successful if it produces practical, inclusive, and implementable outcomes. First, it should ensure meaningful participation of developing countries in shaping global AI governance. Countries such as Bangladesh should not only be consulted, but also be able to influence priorities, standards, and follow-up actions. Participation of civil society, academia, and other non-state stakeholders is equally important. Second, success would require concrete support for national capacity-building. For Bangladesh, this includes AI literacy, regulatory and institutional readiness, research collaboration, digital infrastructure, and technical assistance. Global governance arrangements will be more effective if they take into account differing levels of preparedness across countries. Third, the Dialogue should promote clear safeguards relating to human rights, transparency, accountability, and fairness. This is particularly relevant where AI may affect employment, education, health, public service delivery, and access to information. Fourth, linguistic and local contextual inclusion should be recognized as a governance priority. AI systems should work effectively in Bangla and reflect local realities so that technological progress does not widen existing inequalities. Overall, the Dialogue would be successful if it results in a cooperative framework that is inclusive, rights-based, development-oriented, and responsive to the needs of countries like Bangladesh.

• AI capacity-building
• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Interoperability of governance approaches

Safe, secure and trustworthy AI is critical because AI systems can affect public services, education, health, information access, and employment. Bangladesh needs AI that people can trust, especially where digital literacy and institutional safeguards are still developing. AI capacity-building is equally important because effective governance depends on national readiness. Bangladesh needs stronger technical skills, regulatory capacity, research opportunities, local innovation, and public awareness so that AI governance is not only discussed at the global level but also implemented in practice. Interoperability of governance approaches matters because AI is global, while regulation is often national. Bangladesh would benefit from governance approaches that are flexible enough to reflect local realities, but also compatible with international standards. This can support cross-border cooperation, reduce fragmentation, and help developing countries participate more effectively in global digital governance. Transparency, accountability, and human oversight are essential to protect rights and public interest. In Bangladesh, these principles are especially important where AI may influence decisions related to jobs, finance, education, media, or public administration. People should know when AI is being used, understand how decisions affect them, and have access to human review and remedy where harm occurs. Together, these four areas create a balanced foundation for AI governance. They address both opportunity and risk, while helping ensure that AI remains inclusive, rights-based, and development-oriented for countries like Bangladesh.

First, data governance and data justice should be more explicit. AI governance is not only about model safety, but also about who collects data, whose data is missing, how consent works, how public data is used, and whether people can challenge harmful outcomes. This is especially relevant for Bangladesh, where the draft AI policy and related legal reforms are still evolving. Second, language and cultural inclusion should be treated as a core issue, not a side concern. If AI systems work poorly in Bangla or are trained mainly on foreign contexts, they may exclude large parts of the population and reproduce bias in education, public services, media, and access to information. UNESCO's Bangladesh AI readiness work notes that Bangla-language AI remains limited. Third, there should be stronger attention to labour impacts and a just transition. In Bangladesh, AI governance must consider workers whose jobs may be transformed by automation, especially in sectors linked to garments, services, and routine administrative work. This includes reskilling, social protection, and fair sharing of productivity gains. Global evidence also shows that AI-related job exposure can have gendered effects. Fourth, market concentration and equitable access to AI infrastructure are emerging concerns. A few countries and companies control much of the compute, data, and advanced models. Without addressing this imbalance, global AI governance may remain formally inclusive but materially unequal for countries like Bangladesh. Overall, these issues matter because they shape whether AI governance is truly inclusive, development-oriented, and fair in practice.

In Bangladesh, the main impact of AI governance gaps is that AI adoption is moving faster than the institutions needed to guide it. This creates a mixed picture: strong opportunity for innovation and public service improvement, but also real risks of exclusion, weak accountability, and uneven protection of rights. Bangladesh now has a draft National AI Policy under review, and UNESCO's AI Readiness Assessment has given the country a clearer roadmap for ethical and inclusive AI governance. The most significant challenge is the gap between policy ambition and implementation capacity. Bangladesh still needs stronger regulatory capacity, clearer oversight and redress mechanisms, AI procurement standards, and more institutional ability to ensure transparency, accountability, and human oversight in practice. Data governance also remains a major issue, including consent, privacy, responsible data use, and structured data-sharing frameworks. A second challenge is inclusion. Although mobile access is widespread, only about 45% of the population was using the internet in 2023, which means the benefits of AI may be concentrated among already connected groups. There are also persistent gaps in Bangla-language and minority-language AI, which can limit fairness and accessibility. The biggest opportunity is that Bangladesh is not starting from zero. It already has strong digital public service experience, high trust in government digital platforms, growing AI research, and emerging local efforts to build Bengali-language AI systems. If linked with serious capacity-building, these assets could help Bangladesh use AI in education, public administration, health, agriculture, and citizen services in ways that are locally relevant and socially beneficial. Overall, the key issue for Bangladesh is no longer whether AI will spread, but whether governance can evolve quickly enough to make that spread fair, safe, and development-oriented.

In my view, the AI Dialogue can play a useful role if it becomes the place where international cooperation on AI moves from scattered debate to practical coordination. Because it is being set up as an inclusive UN platform for states and stakeholders, and is hosted by the UN Secretary-General with ITU support, it can help build shared understanding across governments, industry, civil society, and the technical community. Its first role should be to reduce fragmentation. Many countries are developing AI rules separately, with different definitions, safeguards, and levels of capacity. The Dialogue can help identify common principles and areas of interoperability, while still allowing countries to adapt governance to national realities. For countries like Bangladesh, this matters because too much fragmentation increases compliance burdens and can leave developing countries as rule-takers rather than rule-shapers. Its second role should be to connect governance with capacity-building. International cooperation should not only produce norms, but also support implementation through technical assistance, policy learning, research collaboration, and institutional strengthening. Without this, global AI governance risks remaining aspirational for countries with limited resources. Third, the Dialogue can help promote trust and accountability by encouraging exchange on transparency, human oversight, risk management, and public-interest safeguards. It can also amplify perspectives from the Global South, including language inclusion, equitable access to infrastructure, and development priorities that are often underrepresented. Overall, the Dialogue will add value if it becomes not just a discussion forum, but a bridge between global principles and practical cooperation.

The AI Dialogue should build on, not duplicate, the strongest parts of the existing ecosystem. At the global level, it should connect closely with the Global Digital Compact and the Internet Governance Forum, since the UN has already positioned the Dialogue as part of that broader digital governance architecture and as complementary to existing forums. It should also draw on UNESCO's Recommendation on the Ethics of AI and its readiness-assessment work, which already provide a widely accepted normative and capacity-building foundation for Member States. It should also engage with more specialized mechanisms: OECD AI Principles and the integrated GPAI-OECD partnership for policy learning and implementation; ITU's AI for Good platform for technical cooperation, standards, and multi-stakeholder partnerships; ISO/IEC SC 42 for technical standards; and the Council of Europe Framework Convention on AI for human-rights-based legal approaches. These initiatives already cover important parts of the governance landscape, but they remain fragmented across policy, standards, ethics, and sectoral practice. The added value of the AI Dialogue is that it can become a bridging space across these tracks. It can help align principles with implementation, identify gaps and overlaps, promote interoperability between governance approaches, and ensure that capacity-building, financing, and inclusion are treated as central issues rather than side topics. For countries like Bangladesh, this matters because navigating multiple parallel frameworks is difficult; a UN-based Dialogue can bring coherence, legitimacy, and stronger representation of Global South priorities, including language inclusion, equitable access, and public-interest use of AI.

Different stakeholders should contribute in ways that reflect their actual roles and knowledge. Governments can share regulatory experience, policy gaps, and implementation challenges. The private sector can provide technical evidence, risk-management practices, and transparency commitments. Civil society, journalists, workers' representatives, youth, women's groups, disability advocates, and affected communities can show how AI is affecting real people on the ground. Academia and the technical community can contribute independent research, standards expertise, auditing methods, and policy analysis. This matches the UN mandate for a Global Dialogue involving governments and all relevant stakeholders, and the Dialogue is already framed as an inclusive UN platform linked to existing UN processes and the AI for Good summit. For the format, the Dialogue should be hybrid, open, and accessible, so participation does not depend only on travel budgets or diplomatic presence. It should also allow written submissions, remote interventions, and plain-language background papers. A good structure would have three layers. First, regional and virtual consultations before the main meeting, so countries and stakeholders from the Global South can shape the agenda early. Second, the Dialogue itself should use small thematic roundtables rather than only high-level speeches. Third, it should end with a short public summary that captures areas of agreement, open disagreements, and practical follow-up steps. Each session should mix stakeholder groups instead of separating them. There should also be balanced representation by geography, gender, and sector, with dedicated support for underrepresented actors, especially civil society and regulators from developing countries. For Bangladesh and similar countries, this would make the Dialogue more inclusive, more practical, and more likely to produce outcomes that can actually be implemented.

In Bangladesh, the most underrepresented voices are not abstract stakeholder groups, but the people most likely to be affected by AI without having much influence over the rules. These include rural communities, women and girls, informal and low-wage workers such as garment and agricultural workers, persons with disabilities, Bangla and indigenous-language users, small local media and civil-society groups, and local researchers or startups outside major urban centres. This matters because Bangladesh still has clear urban-rural and gender gaps in internet access, while the draft National AI Policy itself highlights limited AI literacy, the need for inclusive access for women, children, and underrepresented groups, and the importance of Bangla and other nationally relevant languages. These voices can be included through design, not symbolism. The AI Dialogue should reserve speaking slots and advisory roles for Global South civil society and community-based actors; provide travel support and strong remote participation options; accept submissions in Bangla and other languages; and organize regional and sector-specific consultations before the main meeting. It should also require evidence from affected communities, not only from governments and large technology companies. For Bangladesh, national preparatory consultations should include worker organizations, teachers, journalists, disability advocates, youth groups, women's organizations, and public-interest technologists. UNESCO's AI ethics framework stresses diversity and inclusiveness, and the UN has already framed the Global Dialogue as an inclusive platform with stakeholder consultations. In short, inclusion should be judged by who can shape the outcomes, not simply by who is invited to attend.

In the Bangladesh context, the most effective formats would be those that move beyond speeches and allow people to work through real problems. A good model would combine small thematic roundtables, scenario-based policy clinics, and regional listening sessions. For example, one session could examine AI in garment work, another in agriculture, another in education or public services, with each group asked to identify risks, safeguards, and practical governance needs. This would make the Dialogue more useful for countries where AI governance is still developing and where local realities differ sharply from those of major AI-producing states. That matters in Bangladesh, where the draft National AI Policy itself points to persistent urban-rural digital divides and calls for AI systems that support Bangla and other nationally relevant languages. I would also recommend multilingual citizen panels, youth and worker assemblies, and "evidence in 5 minutes" sessions where civil society, researchers, journalists, and community representatives present one concrete case of harm, need, or opportunity. For Bangladesh, remote participation should be treated as essential, not optional, because meaningful inclusion cannot depend only on travel, English fluency, or diplomatic access. Simultaneous interpretation, plain-language briefing notes, and written inputs in Bangla should be built in from the start. A strong closing format would be a public synthesis session that clearly records areas of agreement, disagreement, and next steps. This would be more dynamic than standard panels and more credible for underrepresented participants. It also fits Bangladesh's own policy direction, which emphasizes structured public consultation, linguistic inclusion, and transparency in AI governance.

In the Bangladesh context, a few approaches stand out as especially useful. First, Bangladesh's draft National AI Policy 2026-2030 offers several practical tools that could become good governance models if implemented well. These include Algorithmic Impact Assessments for significant and high-risk AI systems, regulatory sandboxes for supervised testing, a proposed central coordination role for the National Data Governance Authority, and use of existing grievance channels for public redress. This is a strong example because it combines innovation support with accountability and risk management. Second, UNESCO's AI Readiness Assessment Methodology (RAM) is a good practice because it does not begin with regulation alone. In Bangladesh, the RAM process brought together government, academia, civil society, and the private sector to assess national readiness and identify priorities for rights-based, inclusive, and evidence-based AI governance. That kind of multi-stakeholder diagnostic approach is valuable for countries still building institutions and capacity. Third, Bangla-language AI infrastructure is itself a governance solution. Platforms and communities such as Bengali.AI, which build open Bangla datasets and language technology resources, help address one of the biggest governance challenges in Bangladesh: exclusion caused by weak local-language support. Governance is not only about restricting harm, but also about ensuring that AI works fairly for local users. Finally, effective AI governance in Bangladesh should build on human-rights, data-protection, and cyber-security frameworks. UNESCO notes that Bangladesh's policy direction already emphasizes rule of law, dignity, human oversight, and alignment with the Personal Data Protection Ordinance, 2025 and the Cyber Security Ordinance. This kind of legal alignment is essential for trustworthy AI.