Putting data justice at the heart of AI governance

Soledad Vogliano argues that data justice goes beyond technical or procedural questions about privacy and interoperability, and must instead interrogate the fundamental power dynamics behind data collection and use. The question is not simply how data is governed, but in whose interest and with what consequences for people and the planet. This framing situates data justice within broader struggles for self-determination and equitable development.

She explicitly states that data justice means asking not only how data is governed, but mainly posing a question about whose interests, under whose control, and with what consequences for peoples and the planet data-driven technologies are being deployed . She further argues that data justice cannot be reduced to abstract understandings of privacy, access, interoperability, or benefit sharing, and must instead address contextualised understandings of power .

Vogliano contends that data justice is fundamentally about power relations — specifically, who holds decision-making authority over data collection, infrastructure, and the benefits and risks of data-driven systems. She frames these as political questions that cannot be resolved through purely technical or regulatory means. This power-centred analysis is presented as essential to achieving a genuinely people-centred information society.

She poses a series of pointed questions to illustrate the power dimensions of data governance: who decides what data is collected, who defines what counts as useful knowledge, who controls the infrastructures where data is stored and processed, who benefits from data-driven innovation, and who carries the risks when data is used to automate decisions, train AI systems, expand surveillance, shape markets, or reorganise access to land, seeds, credit, insurance, and climate finance .

Vogliano argues that the digital transformation of agriculture was not designed to serve the interests of the communities that actually produce the world's food, but rather to expand the market power of agri-industry and big tech. This has resulted in a system that extracts data from rural and indigenous communities while creating dependency on corporate-controlled services. She presents this as a concrete example of how datafication reproduces historical patterns of extraction.

She notes that the architecture of digitalisation and the push for datafication in agricultural ecosystems was not driven by the interests of food sovereignty subjects, but by opportunities seized by agri-industry and its integration with big tech to expand models of production where industry can profit by controlling markets and external inputs . She references a recent report from the panel of experts on sustainable food systems, which stated concern that the currently widely ungoverned ag-tech system is reproducing the logic of extraction of data while creating dependency on corporate-controlled data-driven services .

Vogliano emphasises that the communities most affected by agricultural datafication — including indigenous peoples, pastoralists, fisher folk, and rural women — are not passive data subjects but rights holders, knowledge holders, and political actors. Despite producing the majority of the world's food, these groups were excluded from the design of the digital systems that now govern their livelihoods. This exclusion is presented as a fundamental injustice that data governance frameworks must address.

She states that food systems today impact approximately 3.45 billion people worldwide , and that the relevant subjects include indigenous peoples, pastoralists, fisher folk, agricultural workers, women, and rural communities who produce 70% of the world's food through subsistence, nutrition, variety, and cultural adequacy within models that respect planetary boundaries . She emphasises that these groups are rights holders, knowledge holders, and political actors whose practices sustain food systems, biodiversity, and ecosystem resilience .

Vogliano provides a concrete example of how datafication can undermine hard-won protections for communities in the Global South. By separating digital sequence information about seeds from the physical seed, biotech corporations are able to claim ownership over the data, effectively circumventing the intellectual property protections established under the FAO seed treaty. This illustrates how data governance failures in one arena can have profound material consequences in another.

She explains that food movements have been fighting for 50 years for the protection of seeds, which is central to food sovereignty in Africa, Latin America, and Asia . She describes how the process of separating digital sequencing information about seeds from the materiality of the seed - occurring at the Convention on Biological Diversity (CBD) - allows the appropriation of seed data by biotech corporations, undoing decades of work to protect seeds as a commons under the FAO seed treaty .

Vogliano argues that data governance must be designed to empower communities rather than entrench their dependence on corporate systems. This requires building public and community-controlled data infrastructures and establishing frameworks that treat data as a commons. She also calls for the explicit recognition of the rights of indigenous peoples and peasants as enshrined in international instruments.

She states that for food movements, data governance should support autonomy rather than dependency, strengthen public and community-controlled infrastructures rather than deepen corporate concentration, and promote frameworks for the commons while protecting rights, including the rights of indigenous peoples and peasants as outlined in UNDROP and UNDRIP, rather than relying only on individual consent models that are inadequate in many of these cases .

Vogliano highlights a critical awareness gap: the communities most affected by agricultural datafication, as well as the regulatory institutions meant to protect them, do not adequately understand the implications of these technologies. This gap is compounded by the rapid pace of technological deployment, which outstrips the capacity of both grassroots actors and governance institutions to respond. She presents this as a major obstacle to achieving data justice.

She states that the answer to whether grassroots actors are aware of the relevance of data governance and AI governance conversations is 'not at all', and describes an incredible gap between the speed of development and deployment of technologies and the understanding not just by grassroots actors but by the institutions that actually regulate the sector . She references the Smart Farming Conference at FAO and the high-level forum organised by the Committee of Food Security and Nutrition addressing AI and digital technologies as examples of this lack of understanding .

Vogliano deepens the discussion on linguistic diversity by arguing that the problem extends beyond language to encompass entire worldviews and knowledge systems. Much of the world's knowledge — particularly that held by indigenous and rural communities — exists in oral and non-written forms that AI systems cannot ingest, regardless of translation capabilities. This homogenising tendency of AI represents a fundamental threat to cultural and epistemic diversity.

She argues that it is not just about language but about visions of the world, and that most visions of the world are not systematised in a paper, so even the best translator would not enable AI to capture much of what exists in the world . She states that this should be a baseline assumption about how we think of the interaction of AI and the world .

Vogliano argues that the strategy for advancing data justice must be grounded in cross-sectoral solidarity among movements fighting for justice in concrete, material domains. Because datafication is transforming the very issues these movements care about — such as seed sovereignty and food systems — they must understand data governance as central to their struggles. This requires building connections across movements rather than treating data justice as a separate, technical domain.

She states that the data justice strategy must be supported by the capacity to connect among movements fighting for justice in the real world, not just about data, because data is transforming what they do and what they fight for . She uses the example of seed sovereignty and the appropriation of digital sequence information to illustrate how datafication changes the power relations over systems of production and how people live .

Nandini Chami draws on the development economics tradition of starting from injustice to understand justice, applying this to the global AI landscape. She presents stark statistics on the concentration of AI computing capacity to argue that the current digital order reproduces colonial dynamics. Countries in the Global South are structurally disadvantaged, with intangible assets and innovation accruing to the Global North while costs and low-paid labour remain in the South.

She cites research by the UNU indicating that more than 90% of AI specialised computing capacity is concentrated in just two countries, while over 150 nations lack significant domestic AI infrastructure . She notes that oligopolies characterise all levels of the AI stack from semiconductor manufacturing to frontier model development and cloud infrastructure , and that without infrastructural capabilities, developing countries end up trapped in a new colonial dynamic where intangible assets accrue to the Global North while technological costs, including critical mineral mining, are borne by the Global South .

Chami argues that the concentration of AI capabilities in a small number of corporations creates a structural dependency for developing countries that has severe macroeconomic consequences. As domestic economies rely increasingly on foreign AI models and cloud services, they face rising foreign exchange deficits. This risks creating a two-speed economy in which a hyper-productive, foreign-owned enclave coexists with a suppressed domestic sector.

She references scholar Srinivas Raghavendra's caution that this risks reproducing a dual economy or two-speed structure, where the economy forks into a hyper-productive AI-integrating foreign-owned enclave that sets the national cost base, while the domestic sector is forced to adjust by suppressing wages, operating on thin margins, and under-investing simply to remain viable in export markets .

Chami challenges the dominant narrative that free data flows are universally beneficial for development, arguing that this approach ignores the vastly unequal starting points of different countries. Countries with weaker digital infrastructure, less capital, and less bargaining power are systematically disadvantaged by harmonised free data flow regimes. Imposing a one-size-fits-all solution risks permanently locking in existing inequalities.

She states that the push for a harmonised free data flows regime as the one-size-fits-all solution for development risks permanently entrenching impediments to data justice, as it does not recognise the deep asymmetries in digital infrastructure, capital, and bargaining power that leave countries at different starting points . She also notes that trade and taxation regimes prevent the effective redistribution of data value .

Chami identifies a fundamental contradiction in the intellectual property frameworks governing AI: corporations claim that using others' data to train AI models is permissible, while simultaneously seeking proprietary protection for the outputs of those models. This asymmetry systematically disadvantages data-producing communities and countries. She argues that resolving this requires grounding international economic law and data governance in the principle of sovereign equality of states.

She notes that scholars have flagged that firms claim training models on others' data is permissible, while simultaneously arguing that AI-generated outputs and trained models should receive proprietary protection . She concludes that international economic law and data and AI governance need to be relocated in the sovereign equality of states principle in order to effectively challenge data injustice .

Chami argues that the apparent tension between privacy and open access to data is in large part a product of the commodification of data, and that genuinely protecting privacy requires limiting the extent to which data can be treated as a commercial asset. She contends that digital tools already exist to enable the provision of anonymised data in the commons, but that this option is not pursued because it does not serve the interests of those who hold the most power in the data economy.

She states that we will never be able to successfully protect the right to privacy unless we set limits on data commodification . She argues that advances in digital tools should make it much easier to provide anonymised data in the commons, and that we have the tools to strengthen the commons but are not using them because the commons is not in the interest of those who have most of the power .

Chami uses the example of the African Commission on Human and People's Rights to demonstrate that the relationship between data access and rights protection need not be framed as a zero-sum trade-off. By recognising access to data as a precondition for the right to information, the Commission offers a model for integrating data access into a human rights framework. This reframing presents data access and privacy protection as potentially complementary rather than opposing goals.

She notes with pride that the African Commission on Human and People's Rights passed a resolution two years ago recognising access to data as a precondition for access to the right to information .

Ambassador Kah argues that data governance is not a parallel or subordinate concern to AI governance, but rather its structural foundation. Because every AI system depends on the data flows that data governance frameworks regulate, separating the two conversations is both analytically incorrect and practically harmful. He calls for a formal channel to ensure that data governance findings feed directly into AI governance processes.

He states that data governance is structurally upstream of AI governance, and that every AI system is trained and deployed on the very data flows the working group on data governance is negotiating, making treating these as separate conversations an artificial and costly separation . He describes himself as a 'data evangelist' who raises his hand in AI discussions to remind participants not to forget the foundation that drives AI systems, arguing that this is where the economics, control, justice, equity, trust, safety, ownership, and benefit-sharing issues reside .

Ambassador Kah explains the origins, mandate, and institutional context of the UNCSTD Working Group on Data Governance. The working group was established as a follow-up to the Global Digital Compact and the Pact of the Future, with a specific mandate to develop recommendations on equitable and interoperable data governance. Its work is intended to feed into the UN General Assembly's oversight of global digital governance.

He explains that the working group is a multi-stakeholder body established in 2025 under the United Nations Commission on Science and Technology for Development, tasked as a continuing work from the Pact of the Future and Paragraph 48 of the Annex of the Global Digital Compact . Its mandate is to produce recommendations towards equitable and interoperable data governance arrangements and report progress to the United Nations General Assembly .

Ambassador Kah outlines the four thematic tracks that organise the working group's substantive work, each of which mirrors a key area of the Global Digital Compact. These tracks cover the full spectrum of data governance concerns, from foundational principles to practical questions of interoperability, benefit sharing, and cross-border data flows. Each track has a co-facilitator and has conducted open consultations globally.

He describes the four parallel pillars or tracks: the first is fundamental principles of data governance; the second is interoperability between national, regional, and international data systems; the third is how to share the benefits of data; and the fourth is safe, secure, and trusted data flows including cross-border data flows . He notes that each track has a co-facilitator and has run open calls for input and consultation across the globe .

Ambassador Kah highlights the distinctive multi-stakeholder composition of the working group as a key feature that sets it apart from traditional intergovernmental processes. With equal numbers of state and non-state members, the working group gives civil society, academia, youth, and the technical community a genuine voice in shaping the outcome. He presents this near-parity as essential given the complexity of data governance.

He states that the composition is quite unique, with 27 state members and 27 non-state members, including civil society organisations, academia, youth groups, IEEE, multilateral organisations, and observers . He notes that all members have near-equal footing to design the outcome of the report, and that this is not common in the Commission on Science, Technology for Development, because no one perspective will give us data governance as it is too complex for one side of the equation to govern .

Ambassador Kah argues that data justice cannot be added as an afterthought to governance frameworks that have already been designed without it. Instead, it must be embedded as a foundational design requirement from the very beginning of both data governance and AI governance processes. He specifically emphasises the need to ensure that least developed countries and developing countries are represented by design, not as a concession.

He states that data justice must be a part of the process, built into data governance and AI governance from the onset and not retrofitted later on . He adds that representing LDCs and developing countries must be a design requirement in both the process and not as an afterthought .

Ambassador Kah calls for a structural reform in the architecture of global digital governance to ensure that the work of the data governance working group directly informs the AI governance dialogue. He argues that the current situation, in which the two processes are aware of each other but run in parallel, is insufficient. A formal channel would ensure that data governance insights shape AI governance from the outset.

He states concretely that a formal channel is needed for the working group on data governance's benefit sharing and cross-border data flow findings to feed directly into the global dialogues' agenda-setting, rather than the two processes running in parallel and merely being aware of each other .

Ambassador Kah draws a distinction between privacy, which he defines as the safeguarding of individual rights over protected data, and protection, which he frames as a broader concept encompassing communities, public institutions, and organisations. He argues that trustworthy open systems must strike a delicate balance between openness and protection, ensuring that innovation does not undermine security, equity, or digital sovereignty.

He states that privacy is about safeguarding individual rights over protected data, while protection is much broader, including securing people, communities, public institutions, and organisations from exploitation and cyber threats, particularly in the rise of digital public infrastructure . He notes his service on the UN Advisory Group on Digital Public Infrastructure and Safeguards, and argues that from the Global South, trustworthy open systems must balance openness and protections, and that innovation should not come at the expense of security, equity, and digital sovereignty .

Ambassador Kah argues that multilingualism in AI is not simply a matter of translating interfaces into more languages, but requires ensuring that the full diversity of human languages, cultures, traditional knowledge systems, and oral traditions are represented in the data that trains AI models. He uses the example of Cameroon, which has over 200 languages, to illustrate the scale of this challenge.

He states that multilingualism is not simply about translating interfaces, and that it goes beyond and deeper to ensuring that diverse languages, cultures, traditional knowledge systems, and oral traditions find their way into the knowledge systems represented in data that trains large learning models . He uses the example of Cameroon having over 200 languages to illustrate the diversity that must be accommodated .

Ambassador Kah frames the current moment as historically unprecedented: for the first time, the Global South is participating in the design of rules that will govern the world's most significant resources. He contrasts this with previous instances where rules governing major global resources were designed without the participation of those most affected. He urges all stakeholders to recognise and actively support this opportunity.

He states that most of the rules that govern the most significant resources of the world, when they were designed, the Global South was not in the room, and asks how there can be justice in that context . He describes the current moment as the first time that rules governing lives are being shaped with the Global South around the table, and calls this unprecedented, pointing to the scientific committee's report as evidence of a whole spectrum of highly brilliant and competent individuals across the globe shaping the rules .

Arturo Sanchez-Pineda argues that while open access and open knowledge initiatives were well-intentioned and have genuine benefits, they have inadvertently created conditions that large corporations have exploited. By aggregating freely available data produced by communities — including researchers and academics in the Global South — these corporations have been able to develop and sell products without sharing the benefits with the original data creators. This creates a profound imbalance between those who generate data and those who profit from it.

He describes how researchers in the Global South have been encouraged to publish openly and share their knowledge, but that at some point this data becomes subsidised by large corporations that are able to grab all this data and sell it back . He notes that all this creation of data is creating a big imbalance between people who are able to profit on it versus the people who are creating every single piece of data . He uses the example of Wikipedia as an emblematic case where people eagerly contributed content, only for that content to be aggregated and used in ways they did not control .

Sanchez-Pineda identifies a structural bias in academic publishing that has downstream effects on AI training data: researchers in the Global South have been conditioned to publish in English to reach wider audiences, which means that AI models are trained predominantly on English-language content. He argues that a practical counter-strategy is for researchers to maintain knowledge production in their own languages and cultures, which would both preserve linguistic diversity and ensure that AI models are trained on more diverse data.

He notes that as an academic he was trained to write in English because that would ensure his paper reached a wider audience, and that writing in Spanish would mean fewer colleagues would read or translate it . He argues that keeping research in one's own language and culture would serve multiple purposes: it would keep multiple languages in the data that models gather, and it would also allow the use of citation and digital object identifier systems to ensure accountability and trackability of contributions .

Sanchez-Pineda offers concrete, actionable recommendations for communities seeking to assert greater agency over their data contributions. He argues that by maintaining knowledge production in local languages and using existing tools such as citation systems and digital object identifiers, communities can ensure that their contributions are tracked and recognised. This approach leverages existing infrastructure to build accountability into the data ecosystem.

He recommends keeping citations and digital object identifiers for research produced in one's own community, such as theses and papers about local communities, and using these tools so that AI models can track contributions back to their source . He argues that this kind of accountability and trackability is important because many people feel they are giving their data for free without receiving anything back, and that these systems can be used to ensure recognition for what has been produced in one's own language, culture, and environment .

Speaker 1 reaffirms the foundational WSIS vision of an inclusive, people-centred, and development-oriented digital future, which has guided the process for over two decades. They argue that this vision must now move from reflection to implementation, with the next decade focused on delivering concrete results. The multi-stakeholder spirit that has defined WSIS is presented as the appropriate framework for this implementation phase.

Speaker 1 states that the WSIS vision has been remarkably prescient for the last 20 years, envisioning an inclusive, people-centred, and development-oriented digital future . They emphasise that today has to be more than reflection and must be about implementation, with the next decade focused on delivering results together .

A young audience member, Darren, raises the issue of linguistic inequality in AI systems, noting that the vast majority of the world's languages are not adequately represented in AI training data. He frames this as a question of how different stakeholders — nations, civil society, and youth — can work together to make AI more inclusive of diverse linguistic and cultural communities. He specifically asks how AI can be made more Afrocentric, Asian-centric, and Latin America-centric.

Darren, aged 15 and president of the nonprofit SDGs Connect, notes that AI is very focused on a few languages - the six UN languages and a few other common ones - and that of the over 6,000 languages in the world, perhaps only 2-3% are covered well enough to produce good AI responses . He references a side event hosted at the UN Open Source Week where these issues were discussed .

An audience member who is a lawyer raises the practical tension between privacy protections and the public's right to access judicial decisions and legal information. He argues that in his country, Belgium, GDPR-style protections have made it extremely difficult to access court decisions, even though access to justice and legal information is a fundamental right. He questions whether privacy should be treated as a sacred right that overrides the right to access information about justice.

Didier Cornel, a lawyer from Belgium, describes how it is almost impossible to find the decisions of many courts in Belgium, not for technical reasons but because of questions related to privacy and GDPR . He gives the example of rules requiring politicians to publish information in the official journal, but in a format that makes it impossible to access data about people, such as by preventing copy-pasting .

Anriette Esterhuysen, relaying a comment from online participants, raises the importance of including environmental impact and environmental justice in data governance and AI governance discussions. She acknowledges that this dimension was not adequately addressed during the session and apologises to online participants for not giving them the floor. This serves as a reminder that data and AI governance have material environmental consequences that must be part of the conversation.

She conveys a comment from online participants that environmental impact and environmental justice should not be forgotten in these conversations , and apologises to online participants for not giving them the floor during the session .