Tech Global Institute

As the only AI governance forum with universal UN membership, the first Global Dialogue on AI Governance has a genuinely distinctive opportunity. We consider the first Dialogue a success if it achieves the following four things: 1. The Co-Chairs' Summary documents where genuine convergence exists and where it does not. The risk in any UN process is that contested questions dissolve into ambiguous language that commits no one to anything. The Summary must name disagreements by stakeholder type and region, including minority views and alternative frameworks. Global Majority actors need a record they can point to in order to advance their positions in subsequent processes. 2. The Dialogue must reject the implicit assumption that frameworks from the Global North (such as the EU AI Act or US executive orders) define a baseline from which Global Majority frameworks depart. A law from Brazil, a regulation from Indonesia, or a policy from Kenya should be discussed as equally valid contributions to global AI governance, not as "regional approaches" adapting a standard set elsewhere. 3. A commitment to building contextual AI safety infrastructure for the Global Majority. This means dedicated funding and technical support for multilingual evaluations, culturally grounded benchmarks, and independent, third-party assessments conducted by local research institutions and civil society organizations. The Dialogue should require that any AI safety claim made by developers is accompanied by mandatory disclosure of test conditions, failure rates disaggregated by language and gender, and known limitations. 4. A concrete pilot with a timeline. A successful outcome is an agreement to establish one regional AI audit unit hosted by an existing institution (e.g., a regional IGF or civil society network), funded through existing UN mechanisms, with a mandate to audit two deployed public-sector AI systems (e.g., welfare eligibility, visa screening) before the second Dialogue. The unit must include gender-disaggregated harm reporting as a core requirement.

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

Each theme addresses a different point at which AI governance currently fails Global Majority communities. Governance without capacity defaults to capture. Rights without transparency are unenforceable. Accountability without oversight is a performance. Safe, secure and trustworthy AI: In the Global Majority, AI systems are deployed in high-stakes welfare, healthcare, and law enforcement contexts without evidence of safety or fitness for purpose. The gap is institutional: no regional body in West Africa, South Asia, or the Andean region can independently audit a foreign model before deployment. TGI works with regional partners to document these gaps and advocate for contextual evaluation capacity. Social, economic, ethical, cultural, linguistic and technical implications of AI: Algorithmic discrimination against speakers of under-resourced languages, data labor supply chains that extract value from the Global Majority, and AI-amplified disinformation in electoral processes are largely absent from dominant governance discourse. Also missing are the disproportionate harms faced by women and girls when voice-based systems fail in local languages. These issues must be centered, not treated as regional footnotes to a universal agenda. Protection and promotion of human rights: AI systems are already being used to conduct mass surveillance, deny welfare and identity services through biometric exclusion, and automate discrimination against marginalised communities- with gender-differentiated impacts on widows, women activists, and gender-diverse people. No governance framework built after the fact can substitute for the human rights baseline that already exists and must anchor this Dialogue's work. Transparency, accountability, and human oversight: Systems that remain invisible cannot be challenged by the communities they impact. For oversight to be meaningful, it is essential that affected populations-rather than just developers or regulators-possess the necessary information to comprehend and dispute AI-mediated decisions governing their lives.

Yes. Four critical cross-cutting issues require explicit attention. Power Asymmetries and Data Justice: AI compute, model development, and benchmark datasets are concentrated in a small number of Global North companies and jurisdictions. Data justice questions - who bears the costs of extraction, what conditions data workers face, who benefits - are absent from technical governance discussions. Women constitute the majority of data workers globally, yet are excluded from governance design. Environmental Costs of AI Infrastructure: Accelerating water and electricity consumption from data centers combined with e-waste from rapid hardware cycles, disproportionately affects Global Majority, which hosts a disproportionate share of data centers and raw material extraction operations. Gender-differentiated impacts include women's increased care work burdens when water access is compromised and the exclusion of women from e-waste recycling value chains. Labor and Algorithmic Management: Gig and data workers across the Global Majority face algorithmic wage-setting, performance surveillance, and instant termination with no appeal mechanism. The Dialogue should explicitly address labor conditions, redress mechanisms, and worker protections that are currently absent from governance frameworks. AI in Conflict and Surveillance Contexts: AI-enabled surveillance, autonomous systems, and algorithmic targeting in conflict-affected settings represent a governance gap that falls disproportionately on Global Majority communities. Technology-facilitated gender-based violence using AI-generated content is rising with no adequate governance response. The Dialogue should acknowledge AI-enabled surveillance and targeting as a gap requiring dedicated attention.

Across the Global Majority in South Asia, East & West Africa, and Latin America, AI systems are deployed without local auditing capacity. Harms include algorithmic management without redress, credit scoring that excludes women smallholders, biometric welfare systems without appeal, and surveillance of environmental defenders. The most significant governance gap is structural dependency. The majority of AI compute, foundation models, and evaluation benchmarks are controlled by Global North companies. Global Majority countries import models trained on foreign data, optimised for foreign contexts, deployed without local oversight. Women and girls are disproportionately affected when systems fail in low-resource languages or exclude them through biometric mismatches. At the labour level, data workers and gig workers across the Global Majority power AI systems with no collective bargaining, no appeal rights, and no legal recourse across borders. Women data workers face gender-based wage discrimination and harassment with no reporting mechanism. At the cultural and linguistic level, LLMs accelerate marginalisation of low-resource languages. Indigenous knowledge systems are scraped into training datasets without consent. Women traditional knowledge holders see their expertise appropriated with no recognition or compensation. The Dialogue presents a concrete opportunity to shift the terms of this relationship. Establishing regional AI evaluation centers, modelled on existing regional bodies and funded through the Global Fund on AI proposed under the Global Digital Compact, would allow evaluation of AI systems before deployment in a given context, rather than after harm has occurred. Compute access programs, open benchmarks developed with and for underrepresented languages, and regulatory sandboxes co-designed with local communities would begin to transform dependency into genuine partnership. The goal is to use AI governance as a lever for digital sovereignty, ensuring that Global Majority countries are not permanently relegated to the role of data providers and harm absorbers in a value chain they did not design.

The AI Dialogue can play three essential roles. 1. Tracking Commitments: Most Global Majority countries lack capacity to track the dozen or more parallel governance initiatives. The Dialogue should designate an existing body — the Internet Governance Forum (IGF) secretariat is the obvious candidate — to maintain a public record of what was committed to and what happened, with gender-disaggregated indicators for capacity-building and participation. 2. A Common Governance Vocabulary: The Dialogue can develop a common governance vocabulary that serves Global Majority countries, rather than simply asking them to adopt terminology developed elsewhere. Currently, when Global Majority regulators engage with international AI governance processes, they are largely handed a conceptual framework built around the priorities and contexts of a small number of jurisdictions. A shared vocabulary developed through a universal forum would instead create reference points that allow countries to coordinate across different regulatory traditions without any one framework becoming the default. Without shared reference points, companies can exploit regulatory fragmentation, deploying systems in jurisdictions with the least oversight capacity and the least ability to push back. 3. Democratizing the Conversation. The Dialogue can foster a more open, horizontal discussion that brings together different sectors, countries, and perspectives on more equal terms. Civil society, affected communities, and technical experts from the Global Majority must have a genuine role in shaping outputs, not just presence in the room. The Dialogue's universal membership is its distinctive advantage; this advantage means nothing if the process is designed around actors who already have access and resources.

The Dialogue should build on four existing foundations. 1. The Global Digital Compact (GDC) and its implementation architecture, particularly the Data Governance Working Group under the Commission on Science and Technology for Development (CSTD) and the Global Fund on AI, represent significant commitments already made to Global Majority countries. By leveraging its universal UN membership, the Global Dialogue can hold these international commitments accountable and push for concrete delivery ensuring that global governance initiatives do not remain stalled in high-level rhetoric but instead provide actionable support to less-resourced actors for tangible results. 2. The IGF and its network of National and Regional IGFs offer the most developed existing infrastructure for inclusive, multistakeholder AI governance dialogue at multiple levels. Regional IGFs have spent years building the infrastructure for exactly this kind of locally grounded input. The Dialogue should plug into that network rather than ask communities to engage yet another parallel structure from scratch. 3. The NETmundial+10 multistakeholder guidelines offer a concrete example of how a global technology governance conversation can produce substantive outcomes when participation is genuinely open and the process documents how inputs shaped outputs. 4. The Dialogue should engage with existing civil society networks from the Global Majority for their insights to inform standards-setting and regulatory discussions, rather than treating regional civil society as a consultation checkbox, including the South Asians for Digital Rights Coalition, Global South Alliance, and regional feminist technology networks. These networks bring grounded evidence on how AI systems perform across languages, cultures, and labor contexts.

The Dialogue should be structured around concrete problem-solving rather than sequential statements. 1. Cross-Sector Working Groups: Cross-sector teams each including government, civil society, technical, and private sector representatives, should be tasked with producing a one-page draft recommendation on a specific implementation challenge by the end of the event. Examples might include how to build a multilingual AI evaluation benchmark for low-resource languages, or how to design a redress mechanism for cross-border AI-driven hiring discrimination. This format produces tangible outputs and creates genuine interdependence between stakeholder groups rather than siloed contributions. 2. Access Infrastructure: The Dialogue must provide real-time interpretation into regional languages, low-bandwidth participation options, recorded sessions with synthesised summaries, and structured mechanisms for asynchronous input from those facing visa, funding, or caregiving barriers. Financial support for travel, accommodation, and childcare is a core governance enabler. 3. Intersessional Framework: A lightweight intersessional framework, ideally anchored in the IGF, should track whether commitments made during the first Dialogue are acted upon before the second session. Without this accountability thread, the Dialogue risks being a productive conversation with no follow-through. 4. Protected Civil society Participation: Transparent selection criteria to distinguish grassroots and community-based organisations from well-resourced intermediaries to structurally protect civil society participation. Dedicated participation slots for affected communities, particularly women and gender-diverse people, should be funded and enforced.

Gig and data workers face algorithmic wage-setting, performance surveillance, and instant termination with no appeal or collective bargaining mechanism. Women workers face additional precarity: pregnancy discrimination, gender-based harassment, and exclusion from platform governance. Indigenous peoples face distinct and largely unaddressed risks. The scraping of cultural knowledge and traditional practices into training datasets without consent, the way language models flatten and misrepresent indigenous knowledge systems, and the absence of any meaningful data sovereignty framework that would give communities control over their own knowledge are ongoing harms. Women and gender-diverse people are underrepresented in AI design and governance while bearing the brunt of AI-mediated harms, from biometric exclusion in welfare systems to technology-facilitated gender-based violence. Small island developing states face compound exposure to AI harms, including climate modeling failures and deepfake-driven climate denial, with minimal institutional capacity to respond or participate in the governance conversations where responses are being designed. Non-English speakers and low-resource language communities experience systemic AI failures on contextual accuracy, amplifying misinformation and eroding digital trust. Grassroots civil society organisations are often first to document AI harms but excluded from global norm-setting due to visa barriers, language exclusion, and lack of dedicated funding. Inclusion requires paid fellowships for community representatives; pre-Dialogue consultations whose outputs are presented in plenary; transparent selection criteria preventing well-resourced Northern organisations from filling slots; reverse panels where community representatives question policymakers; and gender-balanced participation quotas with childcare funding.

Four formats would foster meaningful engagement. 1. Session Documentation: Every working group and panel should produce a brief public summary that names specifically where participants agreed and where they did not. This gives Global Majority civil society a record they can reference in national and regional advocacy after the Dialogue ends. Summary documents that smooth over disagreement help only those who prefer the status quo. 2. Direct Testimony Sessions. The Dialogue must create spaces where affected communities, gig workers denied appeal rights, women excluded from biometric welfare systems, Indigenous knowledge holders whose data was scraped without consent, present evidence directly to policymakers. Each session must reserve time for policymakers to respond publicly to what they heard. 3. Design Workshops with Deadlines. Mixed groups (workers, regulators, company representatives, civil society) receive one concrete problem, for example, "Design an appeal mechanism for algorithmic termination that works across three countries," and 90 minutes to produce a one-page recommendation. This forces trade-offs and real exchange, not prepared statements. 4. Accountability Reverse Panels: Community representatives question policymakers and company officials on their actual record: What commitments have you made? What have you delivered? What barriers remain? This shifts power from passive testimony to active accountability. All formats require interpretation, documentation, and clear pathways from recommendations to final outcomes.

Effective AI governance requires concrete practices that make participation meaningful, not symbolic. The NETmundial+10 process demonstrates that transparency about how inputs were handled, including explaining why some inputs were not adopted, builds trust and legitimacy. The Dialogue should adopt the same standard, publishing synthesis reports explaining how each input shaped the Co-Chairs' Summary, including where inputs were not adopted and why. A dedicated online platform must centralize information, publish background materials and guiding questions with adequate advance notice, enable transparent written feedback, and provide synthesized summaries of recorded sessions. Materials must be available in the widest possible range of languages, with low-bandwidth access options. Regional and intersessional consultations are equally critical. The Dialogue should engage with existing regional IGFs, civil society networks to ensure conversations are grounded in locally specific experiences. Regional pre-consultations, conducted in local languages with dedicated funding for grassroots participation, should feed formal positions into the Dialogue rather than serve as preparatory courtesies. By integrating these methods, the Dialogue can shift participation from passive attendance to consequential engagement.