Open Data Collaboratives

The Global Dialogue on AI Governance succeeds if it produces binding, implementable commitments with gender equality as a structural axis. On technical and governance architecture: Mandatory pre-deployment gender impact assessments must become a binding due diligence requirement, with disaggregated data standards built in from the start. The burden of proof must shift from those who experience harm to those who design and deploy the systems that cause it. Interoperability must be a core technical requirement breaking monopolistic lock-in, enabling right to repair, and establishing shared standards for reporting, evidence preservation, and identity protection that place user safety and dignity at the center. On red lines: The Dialogue must operationalize existing calls for moratoria on AI applications that violate international human rights law including systems that enable technology-facilitated gender-based violence, AI-generated non-consensual intimate imagery, biometric surveillance deployed against women and gender-diverse people, and autonomous weapons systems that disproportionately harm civilian women in conflict. Voluntary corporate commitments have failed. Binding regulatory frameworks, enforced by states and supported by UN agencies including UN Women and UNFPA in monitoring implementation, are the minimum threshold. On participation and power: Global South communities and domain practitioners including but not limited to health workers, labour negotiators, and human rights defenders must be logistically and financially supported to articulate what they see is needed to ensure the potential benefits of AI are shaped and shared by all gender-diverse people, and not solely focus on harms that must be prevented. This requires dedicated financing for women's rights organisations to participate as design and delivery partners in this Dialogue and in every process it generates. Representation must have resources to realize impact, and to be part of the development of the governance and design of deployment, never as a consideration after the tools and policies are developed.

• AI capacity-building
• Interoperability of governance approaches
• Open-source software, open data and open AI models
• Transparency, accountability, and human oversight

B. AI Capacity building Institutional pipelines: Fund women into standards-setting bodies (ISO, IEEE, ITU Leadership training: audit AI systems, design evaluation frameworks, set procurement standards, lead governance processes Mentorship and networks. Fund and partner with women-led AI research labs Quotas for Compute E. Interoperability of governance approaches Interoperability must be a core requirement of AI government with the end goal of breaking monopolistic lock-in, enabling the right to repair, and creating shared standards for reporting, evidence preservation, and identity protection. Well-implemented interoperability shifts power away from those who own and profit from AI toward the people it affects. Labour, health, and climate cannot be governed in silos. AI governance that fails to intersect these sectors will fail the people most exposed to its harms. Transparency, accountability and human oversight Voluntary corporate commitments have failed to protect women and gender diverse people from platform-enabled harm. Participation, transparency, accountability, and human oversight must be operating requirements. Binding due diligence obligations, including mandatory pre-deployment gender impact assessments, must become the standard. Clear red lines are equally non-negotiable. In line with recommendations from the UN High Commissioner for Human Rights, applications inherently incompatible with international human rights law must be prohibited. States must enact and enforce regulatory frameworks. Companies must refrain from developing, deploying, or commercializing systems that pose unacceptable risks including deployment in conflict scenarios which disproportionately harm women. Open source software, open data and open AI models Transparency in training data and labor protections for AI workers. Reinvest in open-source communities. Meaningful connectivity; 40% mandatory compute allocation (20% government, 10% academia, 10% civil society); transparency in infrastructure impacts. Women's authorship: recourse in AI-affected decisions, faster funding for gender-focused innovation, leadership quotas, no-fault divorce laws.

As the Gender in Digital Coalition noted in its WSIS+20 submission, feminist and decolonial data governance must be a named requirement, not an implied aspiration. Five cross-cutting issues require explicit inclusion in the Dialogue's thematic architecture. Digital colonialism through gendered data extractivism. AI systems built on data extracted from Global South communities without consent or benefit-sharing reproduce colonial power relations while excluding those communities from governing systems that affect them. Women and gender-diverse people bear disproportionate harm as both subjects of extractive practices and as people locked out of decision-making. Feminist and decolonial data governance grounded in collective rights, community stewardship, and public value must be a named governance requirement. The full AI value chain. The current thematic framing addresses AI systems while largely ignoring hardware, supply chains, data centres, and the labour conditions within them. Women workers in the Global South are concentrated in the most precarious and invisible parts of this chain: data annotation, content moderation, electronic manufacturing. They also bear the severest consequences of resource extraction and climate impact. Addressing this requires new economic models rooted in care for affected communities. Feminist data governance principles. Collective rights, community control, and meaningful informed consent for workers and users are absent from the current framing. These are foundational to whether AI governance serves people or extracts from them. Fragmentation of the AI governance field. Communities working on existential risk, AI safety, participatory AI, and algorithmic accountability are operating in separate languages with separate red lines, increasingly disconnected from the multilateral system where health, labour, climate, and development expertise lives. The Dialogue has a specific and time-limited opportunity to rebuild that connective tissue. TFGBV as a standalone governance concern. Resolution 79/325 contains no explicit entry point for technology-facilitated gender-based violence. Existing AI frameworks need to include TFGBV to ensure prevention and recourse.

AI systems implemented in public services including welfare eligibility algorithms, predictive policing tools, and automated credit scoring, digital public infrastructures, and content moderation in the Global South operate without mandatory gender impact assessments, creating risks of discrimination, surveillance, and exclusion that fall hardest on already marginalised communities. Women human rights defenders and activists working on gender justice are disproportionately targeted by AI-based harassment and surveillance, including documented deepfake campaigns and AI-assisted doxxing across Latin America, Sub-Saharan Africa, and South and Southeast Asia. These risks are absent from most global AI frameworks, highlighting how quickly they are deprioritised when feminist civil society is not adequately represented in agenda-setting. The concentration of AI development in Global North corporations, and the exclusion of Global South communities from early-stage design, means AI systems routinely fail women, girls, and gender-diverse people— including well-documented degradation of safety, health, and legal aid tools in non-English languages— with direct consequences for access to protection and recourse. AI-driven automation is accelerating job displacement in sectors where women are most concentrated globally: clerical work, data entry, garment manufacturing, care-adjacent roles. These sectors and ILO overall must develop frameworks and impose mandatory labour impact assessments or retraining obligations.

The Dialogue has four concrete roles to play in advancing international cooperation on AI governance. Bridge AI and data governance. Formal structural links between AI governance and data governance processes are absent. The Dialogue must create them to ensure that gender-just data governance principles, including feminist approaches to consent, collective rights, and community data sovereignty, are carried across both tracks rather than siloed in neither. Establish cross-border monitoring for AI-related gender-based harms. Women and gender-diverse people, particularly in the Global South, currently have no multilateral recourse when AI-facilitated violence crosses jurisdictions. The Dialogue must advance cross-border monitoring parameters to close this gap. At the same time, any such mechanism must be grounded in international human rights law, subject to independent civil society oversight, and explicitly prohibited from being used as an instrument of transnational repression against the same populations it is designed to protect including WHRDs, journalists, and activists. Make gender equality a tracked outcome, not solely a cross-cutting theme. The tendency to center economic and technological innovation while downplaying gender-related harms is visible across recent global and national AI processes. The Dialogue must resist this by naming gender equality as a specific, reported outcome with indicators, monitoring, and accountability. Labeling it a principle makes it invisible in practice. Advance international convergence on moratoria. The Dialogue should drive convergence around operationalizing moratoria on AI systems that pose significant risks of gender-based harm, including systems that enable or amplify TFGBV. These precautionary measures are essential to prevent irreversible harm and must remain in place until robust, gender-responsive safeguards, accountability frameworks, and access to remedy are effectively established.

Global frameworks The UNESCO Recommendation on the Ethics of AI and its Women4Ethical AI Platform OECD AI Principles The UN Guiding Principles on Business and Human Rights The Global Digital Compact, including the GDC principle on gender equality, secured through sustained feminist civil society advocacy led by the Gender in Digital Coalition, as a baseline that the Dialogue must implement rather than renegotiate. G7 Hiroshima Guiding Principles on AI The Sao Paulo Multistakeholder Guidelines (2024) as a procedural standard for meaningful inclusive participation. The UNESCO Recommendation on the Ethics of AI and its Women4Ethical AI Platform Guiding principles: for law and policy reforms to address TFGBV, by UNFPA and Derechos Digitales, that outline a framework to advance laws that protect, repair, and uphold the rights of survivors of TFGBV. https://www.derechosdigitales.org/en/recursos/guiding-principles-for-law-and-policy-reform-to-address-technology-facilitated-gender-based-violence-towards-a-system-of-accountability/ These initiatives collectively establish a normative foundation across WSIS, GDC, CSW, and UNESCO processes, which the Dialogue should operationalise rather than duplicate.

The structure of the Dialogue will determine whether it produces transformative outcomes or reproduces existing power dynamics. Five recommendations: Formal civil society participation rights. Civil society must have formal speaking rights in plenaries and dedicated representation on expert panels, in accordance with established UN processes, with access to member state briefings and real-time exchange. Predictable timelines. The Dialogue must establish a clear calendar with dedicated submission windows. Smaller feminist civil society organisations and those affected by funding cuts in the Global South cannot participate meaningfully when required to respond on short notice. Reform the AI Scientific Panel. The Panel's mandate and composition must explicitly require gender expertise and Global South representation. Its annual assessments should be subject to structured civil society review before publication ensuring feminist and rights-based knowledge shapes what counts as evidence, not only what counts as response. A multistakeholder advisory group with guaranteed feminist civil society representation must accompany this, providing substantive input rather than ceremonial inclusion. Geography and visa access. Venue selection must account for the limited resources of feminist and gender rights civil society groups and the visa regimes that dictate who can physically be in the room. Geographic inaccessibility is a structural exclusion mechanism. Sequencing as a governance decision. This is a consequential structural choice. If the Dialogue opens with technical presentations and invites civil society to respond, it will reproduce the dynamic that has defined AI governance to date: affected communities positioned to react rather than lead. The alternative is to begin each thematic session with the communities most affected. Ensure a health practitioner defines unsolved problems before a technologist describes what AI could offer. A rights defender should articulate what dignity requires before governance mechanisms are proposed. This approach produces substantively different conversations because it changes what counts as the starting question.

The communities most affected by AI are the least represented in governing it. The resulting harms compound with each governance cycle that proceeds without them. Global South feminist, digital rights, and gender justice civil society organisations are consistently excluded. Resolution 79/325 addresses travel funding, but travel is the narrowest understanding of what participation requires. Core funding, funding for dedicated policy practitioners within civil society, translation, and meaningful inclusion in agenda-setting must also be included. Without these, attendance is not participation. Largely absent from global AI governance discussions: women human rights defenders; LGBTQIA+ organisations; Indigenous communities with distinct data sovereignty interests; informal sector women workers facing AI-driven labour displacement; women in platform-based work subject to algorithmic control and heightened precarity; and communities bearing the territorial consequences of AI infrastructure, including the environmental and resource burdens of data centers and extractive supply chains. The pattern is documented. Gender in Digital Coalition members' direct experience at recent global AI processes, including the AI Impact Summit in India, confirms that gendered algorithmic bias, structural inequality, and the weaponisation of AI against WHRDs are consistently treated as peripheral concerns. They are peripheral because the voices that would centre them are not structurally embedded in governance design. Absence compounds. When women and gender-diverse people from the Global South are not in the rooms where AI governance is shaped, the resulting frameworks mirror their absence making future inclusion structurally harder, not easier. Inclusion requires structural remedy: guaranteed seats, not invitations. Agenda co-design, not consultation after the agenda is set. The Dialogue should establish these as minimum standards and apply them to every process it generates.

Hybrid and flexible time zones must be standard, with written submissions maintained as a formal option. Dedicated thematic sessions on TFGBV and AI, algorithmic bias in foundation models, impacts of unbalanced datasets on health, justice and economic opportunity, gender and the AI value chain, and feminist data governance, using feminist facilitation methods that actively counterbalance power asymmetries in the room. The Dialogue should host sessions and followons outside the Global North to shift geography for these processes. Structured feedback mechanisms between the AI Scientific Panel and civil society, so that the panel's annual assessments reflect feminist and Global South knowledge, not only institutional research networks.

The most instructive governance precedents for AI come from outside the technology sector. The Belmont Report (1978) emerged from a bioethics crisis and succeeded because it brought together scientists, regulators, ethicists, and community representatives starting from values rather than from the technology it sought to govern. It established that those most exposed to harm must shape the terms of protection, not simply be consulted after those terms are set. Elinor Ostrom's work on governing the commons demonstrated that the most durable governance systems are built by the communities closest to the resource, not imposed from above. This is a direct challenge to the assumption that AI governance must be expert-driven in the narrow technical sense. Local knowledge is a condition of its legitimacy. The IPCC model offers a template for translating domain expertise into governance without flattening it, maintaining scientific integrity while remaining legible and actionable to policymakers across very different national contexts. What these examples share: governance that begins from the knowledge of affected communities, holds competing values in productive tension, and builds legitimacy through participation rather than prescription. None of them retrofitted inclusion after the architecture was set. All of them are more durable for it.