Permanent Mission of Bangladesh to the UN, New York

• Success would be measured by concrete deliverables for the Global Majority. Bangladesh would consider the inaugural Dialogue successful if the Co-Chairs' Summary includes: (i) operationalisation of the Global Fund on AI, with pledging modalities, equitable LDC representation in governance, and disbursement priorities agreed without delay
• (ii) a concrete pathway for LDC and graduating-LDC access to AI compute on concessional terms, including regional sovereign or shared facilities supported by the UN system
• (iii) a UN-coordinated initiative on low-resource and indigenous-language datasets, with explicit reference to Bangla, which is spoken by over 280 million people
• (iv) endorsement of open-source models, open weights where appropriate, open evaluation benchmarks and interoperable digital public infrastructure as foundations of inclusive AI
• (v) recognition of South Asian and LDC representation, including Bangladesh, on the Independent International Scientific Panel and its working groups, with composition reflecting geographic, gender, disciplinary and linguistic diversity
• (vi) a clear signal that risk-proportionate, interoperable governance, than export of any single regional model, is the agreed direction of travel
• (vii) transitional arrangements for graduating LDCs in respect of AI-enabling technologies, secured before, graduation in November 2026
• and (viii) a dedicated 'AI and Future of Work' workstream, co-convened by ILO and ITU, with metrics for AI augmentation versus displacement effects linked to SDG 8 monitoring. Equally important is what success is not. We want aspirational text with resources, capacity-building (not framed as charity), or convergence around a regulatory model that considers national capacities. The credibility of the Dialogue with the Global Majority will rest on whether LDCs, SIDS and LLDCs leave Geneva as co-authors of the architecture rather than recipients of decisions taken elsewhere.

• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Open-source software, open data and open AI models

Bangladesh's selection reflects the country priorities. AI capacity-building is the single highest priority: without compute, skills, open resources and technology transfer, all other thematic areas remain aspirational for LDCs. Operationalising the Global Fund on AI, establishing regional compute facilities on concessional terms, and securing technology transfer on mutually agreed terms-including TRIPS transitional flexibilities for graduating LDCs -are essential and time-sensitive. Social, economic, ethical, cultural, linguistic and technical implications matter directly to Bangladesh because of its linguistic diversity (Bangla is spoken by over 280 million people, and over 7,000 languages worldwide remain at risk of digital exclusion), its labour-market exposure (in Bangladesh more than 650,000 freelancers and ICT exports of approximately USD 1.85 billion in FY2024-25), and its climate vulnerability. AI must augment, than displace, workers in ITES/BPO, RMG, agriculture and freelancing, and must be deployed for adaptation, early warning, precision agriculture and disaster response, while transparency on the energy and water footprints of frontier AI is preserved. Interoperability of governance approaches is selected because Bangladesh seeks common terminology, mutual recognition and shared evaluation methodologies-calibrated to national capacities- rather than the extra-territorial application of any single regional model (EU AI Act, Hiroshima Process, ASEAN AI Guide or AU Strategy). A UN/ITU-backed Global AI Regulatory Sandbox would allow LDC startups to test products without facing high-cost compliance regimes prematurely. Open-source software, open data and open AI models are selected as enablers across the other three: they reduce dependency, enable Bangla-language assets to be developed locally, and make digital public infrastructure inclusive. Bangladesh's a2i Programme, BHTPA and the envisaged National AI Centre of Excellence (NAICE) can serve as national anchors. Together these four convert Bangladesh from recipient to credible co-shaper of global AI governance.

Several cross-cutting issues warrant attention. First, the LDC-graduation cliff: Bangladesh graduates from LDC status in November 2026, narrowing TRIPS flexibilities and raising the cost of accessing AI-enabling technologies. The Dialogue should acknowledge graduating LDCs as a distinct cohort and secure transitional arrangements-including concessional compute access for five years post-graduation and preferential licensing of open-weight AI models-before graduation. Second, the climate-AI nexus: AI's energy and water footprints, particularly of frontier models, intersect with climate vulnerability of South Asian and SIDS economies, yet remain inadequately addressed. Transparency requirements on compute-related energy and water use should be a cross-cutting expectation. Third, linguistic equity as digital infrastructure: low-resource and indigenous languages are not a niche cultural concern but a precondition for inclusive AI; Bangla, with over 280 million speakers, is illustratively underserved by frontier models trained on English-dominant corpora. Fourth, labour-market spillovers from AI-enabled automation by multinational enterprises onto developing-country freelancing, ITES/BPO and RMG workforces- a question of decent work that does not fit neatly within any single theme. Fifth, gender-specific harms, particularly AI-generated GBV content and deepfakes targeting women, requiring legal and technical responses across jurisdictions. Sixth, electoral integrity and information ecosystems in democracies with limited fact-checking infrastructure. Seventh, autonomous-system accountability- including AI-generated content and deepfakes-for which Bangladesh's Cyber Safety Act 2026 offers one national approach but which requires interoperable international norms. Finally, an emerging issue: the risk that compliance-heavy regulatory regimes, designed for high-capacity jurisdictions, are applied extra-territorially to LDC firms without commensurate compute, skills or financing- producing de facto exclusion. A UN-anchored Dialogue is uniquely placed to surface these cross-cutting concerns rather than allow them to fall between cluster-shaped silos.

Bangladesh experiences governance gaps along three intersecting axes. Capacity gaps: limited domestic compute, dependence on imported foundation models, and a technical-ethical gap in tertiary AI education identified in the UNESCO Readiness Assessment Methodology report on Bangladesh. Without concessional compute access and structured technology transfer, even sound domestic policy cannot be operationalised. Regulatory gaps: while Bangladesh has gazetted the Personal Data Protection Act 2026 and adopted the Cyber Safety Act 2026- which supersedes the 2023 Act and includes provisions on AI-generated content, deepfakes and autonomous-system accountability-there is no global standard on cross-border data flows that respects national sovereignty while enabling lawful data exchange for AI training, nor agreed minimum technical standards for algorithmic accountability. The proposed National AI Centre of Excellence (NAICE) and the High-Level National AI Advisory Council require resourcing and clear interfaces with international bodies. Equity gaps: extra-territorial application of regional regimes (EU AI Act, Hiroshima Process and others) imposes compliance costs that Bangladeshi startups cannot absorb, and TRIPS flexibilities will narrow on graduation in November 2026. Significant opportunities also exist. Bangladesh's mature digital public infrastructure (DPI) -including identity, payments and service-delivery rails -provides a foundation for inclusive AI deployment. Its Bangla-language NLP, OCR, sentiment-analysis and speech-to-text assets, developed through EBLICT and a2i, are contributions to the global AI commons. Its 650,000-plus freelancer base and ICT export sector (approximately USD 1.85 billion in FY2024–25, target USD 2.6 billion) position it as a stakeholder in the future-of-work agenda. Climate-adaptation, agriculture, health and skills pilots can scale on open-source models and DPI rails. The most significant challenge is converting these assets into co-shaper status before graduation; the most significant opportunity is the Dialogue itself as the platform on which to do so.

• The AI Dialogue should anchor international cooperation in a universal, inclusive UN framework and act as the multilateral counterweight to fragmentation arising from unilateral, bloc-led or extra-territorial initiatives. Three roles are particularly important. First, agenda-setting on equity. The Dialogue is the only standing forum in which LDCs, SIDS and LLDCs participate as co-authors rather than observers. It should keep the Global Fund on AI, concessional compute access, low-resource language datasets, and TRIPS-related transitional flexibilities at the centre of the cooperation agenda -not as 'capacity-building' annexes but as core deliverables. Second, interoperability without uniformity. The Dialogue should converge governance approaches around common terminology, mutual recognition, shared evaluation methodologies and risk-proportionate registration of high-risk AI systems, while explicitly resisting the export of any single regional regulatory model. A UN/ITU-backed Global AI Regulatory Sandbox Framework would allow LDC startups to test products in a supervised environment without premature exposure to high-cost compliance regimes
• Bangladesh's a2i Programme and BHTPA can serve as national anchors. Third, follow-up architecture and metrics. The Dialogue should establish workstreams that translate cooperation into measurable outcomes - including a dedicated 'AI and Future of Work' workstream co-convened by ILO and ITU, with metrics for augmentation versus displacement linked to SDG 8 monitoring
• an Independent International Scientific Panel with geographic, gender, disciplinary and linguistic diversity
• and standing channels with OHCHR, UNESCO, ITU, UNDP, UNICEF and the WTO on questions that cut across mandates. A credible Dialogue is one whose outputs are visible in domestic AI policy, in regional cooperation, and in the budgets of multilateral institutions. Bangladesh sees its role as a bridge between the Global Majority's equity demands and the technical communities driving AI governance, drawing on its DPI maturity, Bangla-language assets and climate-adaptation experience.

• The Dialogue should build on, not duplicate, existing UN and multilateral instruments while adding a coordinating spine. Foundational instruments include the Pact for the Future and its annexed Global Digital Compact, the UNESCO Recommendation on the Ethics of AI (2021), the OHCHR B-Tech guidance and the UN Guiding Principles on Business and Human Rights, the ITU AI for Good Global Summit, the UN Secretary-General's AI Advisory Body work, and the 2030 Agenda - particularly SDGs 8, 9, 10, 16 and 17. Sector mandates from ILO (future of work), WHO (health AI), WIPO (IP and AI), WTO (trade implications) and UNCTAD (digital economy and LDC graduation) should be drawn into a coherent workstream architecture. Regional and plurilateral processes - the Hiroshima AI Process, the Council of Europe Framework Convention, the African Union Continental AI Strategy, the ASEAN Guide on AI Governance and Ethics, and the EU AI Act - offer lessons but should not be exported as global templates. The Dialogue should also connect to UNESCO's Readiness Assessment Methodology, which has produced a country report on Bangladesh, and to the Internet Governance Forum and WSIS+20 review processes. Bangladesh's specific contributions include: Aspire to Innovate (a2i) and the Bangladesh Hi-Tech Park Authority (BHTPA) as candidate national anchors for sandbox and pilot work
• the Bangla NLP, OCR, sentiment-analysis and speech-to-text assets developed through EBLICT/a2i as contributions to global digital public goods
• the gazetted Personal Data Protection Act 2026 and the Cyber Safety Act 2026 as reference points for national legislative frameworks. The added value of the AI Dialogue is its convening authority across these instruments and its ability to anchor equity, capacity-building and risk-proportionate interoperability as cross-cutting expectations -turning a fragmented landscape into a sequenced, monitored, UN-anchored agenda accountable to the Global Majority.

• Format and structure should reflect that LDCs, SIDS and LLDCs are co-authors, not observers. Bangladesh recommends: (i) Multi-stakeholder by design. Governments, the UN system, the technical and standards community, academia, the private sector and civil society -including women's, indigenous and persons-with-disabilities organisations - should be integrated into preparatory consultations, plenaries and working groups, with clear and equitable speaking time. (ii) Regionally balanced co-leadership. Co-Chairs and panel rapporteurs should reflect geographic, gender and developmental diversity
• South Asian and LDC representation should be visible across leadership roles, not only in audience seats. (iii) Sequenced workstreams between plenaries. Inter-sessional working groups on the Global Fund on AI, compute access, low-resource languages, the Independent International Scientific Panel, AI and the future of work, and the proposed UN/ITU Global AI Regulatory Sandbox should produce concrete inputs to each Dialogue, with published terms of reference and reporting timelines. (iv) Hybrid modalities. In-person plenaries (Geneva 2026, New York 2027) should be supplemented by hybrid regional consultations, virtual rapporteur sessions and asynchronous written inputs to reduce travel-cost barriers for LDC delegations. (v) Transparency and traceability. Submissions, summaries, draft outcomes and revisions should be publicly accessible
• Co-Chairs' Summaries should reference specific national submissions to demonstrate that contributions shape outcomes. (vi) Capacity support for participation. Travel funding, technical briefings and language interpretation-including for low-resource languages where feasible - for LDC delegations and underrepresented stakeholders. (vii) Clear metrics. Each Dialogue should adopt and publish a small set of measurable success indicators against which subsequent editions are assessed. Bangladesh would welcome co-convening or hosting an inter-sessional consultation on linguistic inclusion or on the future-of-work workstream, leveraging its a2i, BHTPA and academic networks.

• Several constituencies remain underrepresented. Least Developed Countries, Small Island Developing States and Landlocked Developing Countries: although nominally included, they are rarely co-authors of agenda or outcomes, and their structural constraints- compute scarcity, narrowing TRIPS flexibilities on graduation, climate vulnerability- are treated as side issues rather than central design parameters. Speakers of low-resource and indigenous languages: Bangla, spoken by over 280 million people, and over 7,000 other languages remain at risk of digital exclusion as frontier models concentrate on English and a handful of high-resource languages. Workers in ITES/BPO, RMG, agriculture and freelancing in developing countries: the people most exposed to AI-enabled automation by multinational enterprises rarely have a seat at AI-governance tables
• Bangladesh's 650,000-plus freelancer community is illustrative. Women, ethnic and religious minorities, indigenous peoples and persons with disabilities: targeted protections and lived-experience input on GBV-related deepfakes, accessibility and bias remain underweighted. Climate-vulnerable communities in delta and coastal regions: directly affected by AI deployments in early warning, agriculture and disaster response. National Cybersecurity Agencies, data protection authorities and ICT Ministries from LDCs: who must implement governance domestically. Inclusion mechanisms include reserved speaking slots and rapporteur roles
• targeted travel funding and interpretation
• co-design of working-group terms of reference with affected communities
• partnerships with civil-society networks, trade unions and indigenous organisations
• and structural inputs from Independent National Human Rights Institutions and Special Procedures mandate-holders. Bangladesh proposes that the Dialogue institutionalise a 'Global Majority Caucus' as a standing consultative mechanism, anchored in the LDC, SIDS and LLDC groupings and connected to UN regional commissions, to ensure that underrepresented voices are not invited episodically but built into the architecture of the Dialogue itself.

Bangladesh recommends formats designed to surface diverse perspectives and produce traceable outcomes rather than statements of position. (i) Pre-Dialogue regional consultations. Hybrid regional sessions hosted by UN regional commissions (ESCAP, ECA, ECLAC, ECE, ESCWA) in the months preceding each Dialogue, feeding written summaries into the plenary agenda. (ii) Thematic interactive panels with rapporteurship. Each cluster anchored by a panel that mixes governments, technical experts, affected communities and youth, with named rapporteurs producing real-time syntheses visible to participants. (iii) Country-pair dialogues. Structured exchanges pairing an LDC, SIDS or LLDC delegation with a high-capacity counterpart on a specific question - compute access, language datasets, sandbox design - to produce joint inputs. (iv) Sandbox demonstrations. Live demonstrations of national AI sandboxes (Bangladesh's a2i and BHTPA can offer one) and Bangla-language NLP, speech-to-text and OCR assets, anchoring discussion in working systems rather than abstract principles. (v) Civil-society and youth assemblies. Parallel structured assemblies whose conclusions are formally received into the plenary, modelled on practices from the Internet Governance Forum and the High-Level Political Forum. (vi) Worker and freelancer voice sessions. Direct testimony from ITES/BPO, agricultural and freelancer communities on AI augmentation versus displacement, feeding the proposed AI and Future of Work workstream. (vii) Open and trackable submissions. A live, public registry of written inputs and the outcome documents that cite them, so contributors can see how their inputs shape the text. (viii) Linguistic inclusion. Interpretation in UN languages plus selected low-resource languages where feasible, and machine-translation pilots transparent about limitations. (ix) Asynchronous deliberation tools. Moderated online deliberation between sessions to extend participation beyond those able to attend in person. (x) Outcome-oriented metrics. Each Dialogue concludes with a published scorecard of deliverables agreed and pending, ensuring continuity from Geneva 2026 to New York 2027 and beyond.

Bangladesh offers the following examples of policies, practices and platforms relevant to effective AI governance. (i) National strategy and institutions. The Draft National Artificial Intelligence Policy 2026-2030 (V2.0) of the ICT Division, the envisaged National AI Centre of Excellence (NAICE) and the High-Level National AI Advisory Council provide a domestic governance architecture aligned with risk-proportionate, interoperable principles. (ii) Legislative foundations. The gazetted Personal Data Protection Act 2026 (PDPA 2026) anchors data protection in domestic law, while the Cyber Safety Act 2026 - which supersedes the Cyber Security Act 2023 - includes provisions on AI-generated content, deepfakes and autonomous-system accountability. These offer reference points for national legislative frameworks without mandating conformity with any single external model. (iii) Digital public infrastructure (DPI). Bangladesh's identity, payments and service-delivery DPI rails offer a foundation for inclusive AI deployment in health, agriculture, education and social protection, illustrating that AI governance and DPI maturity are mutually reinforcing. (iv) Aspire to Innovate (a2i). A flagship platform for digital service delivery, sandbox-style innovation and language-technology development that can serve as a national anchor for a UN/ITU-backed Global AI Regulatory Sandbox. (v) Bangla-language assets. NLP, OCR, sentiment-analysis and speech-to-text resources developed through EBLICT and a2i are candidate contributions to global digital public goods, addressing exclusion of low-resource languages. (vi) Bangladesh Hi-Tech Park Authority (BHTPA). Provides physical and regulatory infrastructure for AI start-ups, suitable as a sandbox host. (vii) Climate-AI applications. Operational deployments in early warning, precision agriculture and disaster response, drawing on Bangladesh's climate-adaptation experience. (viii) Inter-ministerial coordination. An MoFA-chaired mechanism convening the ICT Division, a2i, the Cabinet Division, BTRC, the National Cybersecurity Agency, BCC, BASIS, BACCO and academia ensures coherence between domestic implementation and international engagement -a model adaptable to other developing-country contexts.