UNESCO Women for Ethical AI

Success cannot be measured by the elegance of the communiqué issued at the end of the Dialogue. It must be measured by what changes — structurally, institutionally, and in the lived realities of those most affected by AI systems yet least represented in the rooms where decisions are made. A successful first Global Dialogue would produce, at minimum, three concrete outcomes. First, a clear and enforceable accountability framework — not a set of voluntary principles that dissolve under commercial pressure, but binding commitments with named obligations, timelines, and consequences. The world has enough AI ethics declarations. What it needs are mechanisms with teeth. Second, meaningful structural inclusion of the Global South — not as beneficiaries of AI development, but as co-architects of its governance. This means dedicated seats, resources, and decision-making power for nations in Africa, Southeast Asia, Latin America, and the Pacific, whose populations will bear disproportionate risk from ungoverned AI deployment while having had the least voice in shaping its rules. Third, a standing multilateral body with a clear mandate — one that does not duplicate existing structures but fills the gap between aspirational frameworks and operational accountability. This body must include civil society, technical experts, and affected communities, not only state actors and industry. A Dialogue that produces only consensus language without structure, inclusion without power, or principles without enforcement will have succeeded only in appearance. The first Global Dialogue on AI Governance will be remembered as a turning point — or as a missed one. The difference lies in the willingness to prioritize justice over comfort.

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

The four thematic areas I selected - Safe, Secure and Trustworthy AI; Social, Economic, Ethical, Cultural, Linguistic and Technical Implications of AI; Protection and Promotion of Human Rights; and Transparency, Accountability, and Human Oversight - are not independent priorities. They form an interconnected framework without which AI governance remains aspirational rather than operational. Safe, secure, and trustworthy AI is foundational. Systems that are opaque, poorly validated, or deployed without adequate testing cause measurable harm - disproportionately to vulnerable populations who have the least recourse when things go wrong. Safety is not a feature to be added later; it must be designed in from the start. The social, economic, ethical, cultural, and linguistic implications of AI are where governance most often fails. Technical frameworks built in the Global North frequently erase the linguistic diversity, cultural context, and economic realities of the majority world. AI that does not see you - your language, your face, your livelihood - cannot serve you. This thematic area demands that governance be built from lived experience, not only from benchmark datasets. Human rights protection and promotion grounds everything else in enforceable norms. AI governance without human rights as its non-negotiable floor risks becoming an instrument of the very harms it claims to prevent - surveillance, discrimination, displacement, and the erosion of dignity. Transparency, accountability, and human oversight are the mechanisms by which the other three are enforced. Principles without accountability structures are wishes. This area insists that someone must answer - to regulators, to communities, to history - when AI systems cause harm. Together, these four priorities represent a coherent, rights-centered, human-centered governance agenda that cannot be deferred.

Several critical issues fall between or beneath the listed thematic areas and risk being left unaddressed if governance frameworks do not name them explicitly. Neurotechnology and cognitive sovereignty. Brain-computer interfaces, affective computing, and neural data collection are advancing faster than any existing governance framework can accommodate. The listed themes do not adequately address the unique risks of AI systems that access, infer, or manipulate cognition and emotion. Cognitive sovereignty - the right of individuals to the privacy and integrity of their own mental processes - must be recognized as a distinct human rights frontier. AI and grief, trauma, and death. Generative AI is increasingly used to simulate deceased individuals, provide grief companionship, and interact with people in acute psychological distress. These applications raise profound ethical questions about dignity, consent, and exploitation that cross cultural and religious lines in ways the current thematic structure does not capture. Linguistic and epistemic justice. While the second theme touches on cultural and linguistic implications, it does not go far enough. The overwhelming dominance of English-language training data, Western epistemological frameworks, and Global North problem definitions shapes what AI systems know, value, and recommend. Governance must explicitly address whose knowledge is encoded - and whose is erased - in the systems now being deployed globally. AI in conflict and humanitarian contexts. The use of AI in active conflict zones, for targeting, surveillance, and information warfare, is accelerating. Existing humanitarian law frameworks were not designed for autonomous or semi-autonomous systems. This gap is not a future concern - it is present and urgent. These issues require dedicated attention, not absorption into existing categories where they will remain peripheral.

I speak from two intersecting vantage points: as a Filipino-American professional working in medical device human factors and regulatory compliance, and as an advocate for Global South inclusion in AI governance through UNESCO Women for Ethical AI. In the medical device and healthcare sector, the governance gap is acute. AI-enabled diagnostic tools, clinical decision support systems, and patient monitoring technologies are being deployed in healthcare settings — including in the Philippines and across Southeast Asia — without adequate regulatory frameworks, usability validation standards, or human factors requirements tailored to local contexts. The FDA's evolving guidance on AI/ML-based software as a medical device remains largely inaccessible to regulators and manufacturers in lower-resource settings. Patients in these environments bear the highest risk from ungoverned AI in healthcare while having the least institutional protection. In the Philippines and the broader Asia-Pacific region, the challenges compound. Digital infrastructure gaps mean that AI governance conversations assume connectivity, data literacy, and institutional capacity that do not exist uniformly across the archipelago or the region. Linguistic diversity — the Philippines alone has over 180 languages — renders most AI systems culturally and communicatively inadequate for meaningful deployment. Meanwhile, the opportunity to leapfrog legacy systems using well-governed AI in health, education, and public administration is real but unrealized. The most significant opportunity lies in building governance frameworks that treat the Global South not as a recipient of AI development but as a co-designer of its rules. Regional bodies, civil society networks, and women-led technical communities are already doing this work. The gap is not in vision — it is in resources, representation, and institutional will to include these voices at the table where binding decisions are made.

The AI Dialogue arrives at a moment when international cooperation on technology governance is both more necessary and more fragile than at any previous point in the digital era. Geopolitical fragmentation, competing regulatory philosophies, and the accelerating pace of AI development have created a window that will not remain open indefinitely. The Dialogue must use it deliberately. Its most irreplaceable role is legitimacy-building. No single nation, bloc, or institution commands the trust of the entire international community on AI. The UN system, for all its imperfections, remains the only multilateral architecture with the universality and normative authority to convene genuinely inclusive governance conversations. The Dialogue must leverage this — not squander it on lowest-common-denominator consensus language. It can serve as a translation layer between the technical and the political, between the global and the local, and between the powerful and the marginalized. Effective international cooperation requires shared vocabulary, shared threat assessments, and shared understanding of what is at stake for different communities. The Dialogue is uniquely positioned to build this common ground without flattening the differences that must inform governance design. It can establish the architecture for ongoing cooperation — not as a one-time convening but as the foundation for a standing multilateral mechanism with a clear mandate, inclusive membership, and accountability to affected populations rather than only to member states and industry actors. Most critically, it can normalize the expectation that AI governance is a matter of international law and human rights — not merely of national industrial policy or voluntary corporate commitment. That normative shift, if the Dialogue achieves it, would be its most durable contribution to international cooperation. The Dialogue's role is to make cooperation not only possible, but expected.

A productive AI Dialogue does not begin from zero. A rich ecosystem of initiatives, frameworks, and partnerships already exists — the imperative is connection, not duplication. UNESCO's Recommendation on the Ethics of AI (2021) remains the most comprehensive intergovernmental framework on AI ethics, with 193 member states as signatories. UNESCO's Women for Ethical AI initiative further advances gender-responsive governance and Global South inclusion. The Dialogue should treat these not as background documents but as active foundations — adopting their principles, extending their reach, and funding their implementation in lower-resource contexts. The OECD AI Principles and the Global Partnership on AI (GPAI) have produced substantive technical and policy work, but their membership skews heavily toward high-income nations. The Dialogue can add value by bridging GPAI's technical outputs with the broader UN membership, ensuring that governance standards developed in these forums are stress-tested against Global South realities before being adopted as international norms. The ITU's AI for Good platform and regional bodies such as the African Union's AI continental strategy and ASEAN's Guide on AI Governance represent governance work already rooted in regional context. The Dialogue should amplify these rather than supersede them. The EU AI Act and emerging national frameworks in Brazil, India, and Kenya represent a growing body of regulatory experience. The Dialogue can serve as the comparative learning forum where these approaches are examined for interoperability, equity implications, and transferability. The added value the Dialogue uniquely brings is universality with accountability — the ability to convene all nations, center human rights as a non-negotiable floor, and produce outcomes that carry the normative weight of the UN system. No existing initiative can do all three simultaneously.

Meaningful multi-stakeholder participation is not achieved by issuing open invitations and calling the result inclusive. It requires deliberate design — of who is in the room, how they speak, what weight their contributions carry, and whether the process is structured to surface dissent as well as consensus. Governments must come prepared to make binding commitments, not merely to exchange positions. Their role is not to represent industry interests under a sovereign flag but to speak for the populations — including the most vulnerable — whose lives AI systems are already shaping. Civil society and community organizations — particularly those rooted in the Global South, in indigenous communities, and in populations historically excluded from technology governance — must be resourced, not merely invited. Participation without funding, translation, and institutional support is participation in name only. Technical experts and researchers should contribute through structured mechanisms that translate complex findings into policy-relevant language — and that hold space for dissenting scientific voices, not only those affiliated with well-resourced institutions or industry-funded research. The private sector has a role, but it must be defined and bounded. Industry input on technical feasibility and implementation is valuable. Industry capture of governance outcomes is not. Clear conflict-of-interest protocols are essential. On format and structure, the Dialogue should move beyond plenary statements toward deliberative formats — facilitated working sessions, regional caucuses, and issue-specific technical consultations — that produce concrete outputs rather than recorded positions. Outcomes must be documented with attribution, reviewed against stated commitments, and revisited at subsequent sessions. A single Dialogue that produces no follow-on mechanism will not advance governance. The structure itself must encode continuity, accountability, and the expectation of return.

The architecture of global AI governance has been built largely by and for a narrow demographic: predominantly male, predominantly Western, predominantly affiliated with well-resourced academic or industry institutions. The consequences of this are not incidental — they are structural. Who designs governance determines whose harms are visible, whose rights are protected, and whose futures are considered. Women remain systematically underrepresented at every level of AI governance — in technical research, in regulatory bodies, in multilateral forums, and in the corporate structures that shape AI development. This is not a pipeline problem. It is a power problem. Women are not absent from expertise; they are excluded from the tables where expertise is converted into policy. Initiatives like UNESCO's Women for Ethical AI exist precisely because this exclusion is deliberate enough to require deliberate correction. Women from the Global South face compounded exclusion — by gender, by geography, by language, and by resource constraints that make sustained participation in international forums prohibitive. A Filipino woman working in AI ethics, a Kenyan researcher in algorithmic fairness, a Brazilian indigenous rights advocate navigating automated systems — these voices carry knowledge that no amount of proxy representation can substitute. Racial and ethnic minorities, disabled communities, LGBTQ+ populations, migrant workers, and refugees are among those whose interactions with AI systems are highest-risk and whose governance participation is lowest. Inclusion requires more than invitation. It requires funded participation, translation and interpretation services, flexible and hybrid engagement formats, and governance structures that give these communities decision-making authority — not only advisory roles. The Dialogue must measure its inclusivity not by who was invited, but by who shaped the outcomes.

Traditional UN conference formats — plenary statements, side events, and negotiated communiqués — were designed for a different era of international diplomacy. They privilege those already fluent in multilateral process, reward prepared positions over genuine dialogue, and produce documents that rarely reflect the complexity of what was actually said in the room. The AI Dialogue must do better. Deliberative dialogue formats — structured facilitated conversations where participants respond to one another rather than deliver prepared remarks — have demonstrated effectiveness in complex multi-stakeholder settings. These create conditions for genuine exchange, surface disagreement productively, and build the shared understanding that durable governance requires. Participatory scenario exercises that present concrete, near-future AI governance dilemmas — a healthcare AI misdiagnosis in a low-resource setting, an algorithmic hiring system discriminating against a linguistic minority, a grief chatbot deployed without consent frameworks — ground abstract principles in human consequence and make governance stakes viscerally legible across expertise levels. Regional pre-dialogues, conducted in local languages and rooted in regional AI realities, should feed structured inputs into the global forum rather than asking Global South participants to simply respond to agendas set elsewhere. The global Dialogue should synthesize regional voices, not override them. Digital and asynchronous participation mechanisms — not merely livestreams, but structured input portals with genuine synthesis and response commitments — can extend meaningful engagement to those for whom travel and real-time participation are inaccessible. Youth and intergenerational panels that hold decision-makers accountable to long-term consequences, not only present political calculus, introduce a temporal dimension that governance conversations too often ignore. Format is not neutral. How the Dialogue is structured determines whose intelligence it captures — and whose it wastes.

Effective AI governance already exists in fragments across sectors, jurisdictions, and communities. The challenge is not invention - it is recognition, resourcing, and scaling of what is already working. The EU AI Act represents the most comprehensive binding regulatory framework to date, introducing risk-based classification, conformity assessment requirements, and prohibited use categories. Its extraterritorial reach - applying to any system deployed within the EU regardless of origin - establishes a de facto global floor that has already influenced regulatory thinking in Brazil, Canada, and beyond. Its human factors and transparency requirements for high-risk systems offer a replicable model. The FDA's regulatory framework for AI/ML-based Software as a Medical Device demonstrates how sector-specific governance can operationalize safety and accountability in high-stakes AI deployment. Its iterative approach to predetermined change control plans acknowledges that AI systems evolve - and that governance must evolve with them. This adaptive regulatory model has direct applicability beyond healthcare. UNESCO's Readiness Assessment Methodology (RAM) gives governments - particularly in the Global South - a structured tool for evaluating their institutional, legal, and technical capacity to implement AI ethics frameworks. It is one of the few governance tools designed with lower-resource contexts explicitly in mind. Community-led algorithmic auditing initiatives, such as those emerging from civil society organizations in Kenya, Brazil, and the Philippines, demonstrate that accountability does not require state infrastructure - it requires informed, resourced communities with access to meaningful redress mechanisms. Gender-responsive AI impact assessments, piloted by several UNESCO member states, integrate intersectional analysis into procurement and deployment decisions - ensuring that governance asks not only whether a system works, but for whom, and at whose expense.