University of Oxford

The first Global Dialogue would be successful if it translates existing high-level commitments on safe, inclusive, and rights-respecting AI into more concrete areas of governance focus, while addressing gaps that remain under-integrated in current policy discussions. One such gap concerns the human labour that underpins AI systems. A growing body of research has documented how AI development depends on a large, globally distributed workforce engaged in data annotation, content moderation, and related tasks, often through complex and relatively opaque supply chains. While this dimension is increasingly recognised in academic and some policy circles, it is not yet systematically incorporated into AI governance frameworks. In this context, a meaningful outcome of the Dialogue would be to situate these labour dynamics more clearly within ongoing discussions on human rights, accountability, and AI governance. This could include: Recognising data workers as part of AI value chains, and therefore within the scope of governance efforts Encouraging greater transparency around data production processes, including labour arrangements and sourcing practices Identifying areas where existing international standards (e.g. human rights and labour frameworks such as the Fairwork AI standards) can be more explicitly applied to AI supply chains Opening a sustained line of inquiry within the Dialogue process on how labour considerations can be integrated into AI governance. Addressing this dimension would complement existing work on downstream risks and impacts, and contribute to a more complete understanding of how AI systems are developed and governed in practice.

• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These priorities reflect the need to better account for the full range of social and economic processes through which AI systems are developed and deployed. A growing body of empirical research highlights that AI systems rely on globally distributed forms of digital labour, including data annotation and content moderation. These activities raise important human rights considerations, particularly in relation to working conditions, exposure to harmful content, and access to remedy. At the same time, transparency and accountability remain limited with respect to how data is produced. AI supply chains are often characterised by multiple layers of outsourcing, making it difficult to assess labour practices or assign responsibility. Strengthening transparency in this area is therefore a prerequisite for meaningful oversight. The social and economic implications of AI extend beyond questions of automation and productivity to include the expansion of digital labour markets. These developments can generate new forms of work and income, but also reproduce existing inequalities and precarity. Finally, capacity-building should include not only technical capabilities, but also the institutional and regulatory capacity to understand and govern these labour dynamics, particularly in countries where such work is concentrated.

An important cross-cutting issue is the governance of AI supply chains, particularly the labour conditions under which training data is generated. While existing thematic areas address human rights, accountability, and transparency, they do not explicitly engage with the production processes that underpin AI systems. Research has shown that these processes are often organised through globally distributed and relatively opaque labour arrangements, mediated by subcontracting networks. Two interrelated gaps follow from this: Limited visibility into labour practices within AI supply chains, including how work is organised, compensated, and managed A disconnect between data governance and labour governance, with most discussions of data focusing on quality, bias, or provenance, rather than the human work involved in producing it. This creates a situation in which AI systems can be assessed against technical or ethical criteria, while the conditions of their production remain less scrutinised. Addressing this issue would not necessarily require a separate thematic track, but rather greater integration of supply chain and labour considerations across existing areas, particularly those relating to human rights, accountability, and transparency.

In the context of AI data work and digital labour, governance gaps are directly shaping how value and risk are distributed across AI supply chains. A central challenge is the persistent opacity of these supply chains. Data workers are typically engaged through multi-layered subcontracting arrangements or digital labour platforms, which fragment responsibility and limit oversight. As a result, existing regulatory frameworks, whether on labour, human rights, or corporate accountability, are difficult to apply in practice. This opacity is closely linked to uneven working conditions, including low and unstable pay, limited job security, and, in some cases, exposure to harmful content without adequate safeguards. While such issues are increasingly documented, they are not yet systematically addressed within AI governance frameworks. At the same time, the expansion of AI-related data work is creating new forms of labour market participation, particularly in parts of the Global South. These opportunities are often framed in terms of digital inclusion and economic development. However, in the absence of clearer standards (such as Fairwork as similar standards) and accountability mechanisms, much of the value generated in these supply chains remains concentrated elsewhere, while risks are externalised to workers. Recent advances in AI governance, particularly around human rights, due diligence, and transparency, create an opening to address these dynamics more directly. The key challenge is ensuring that these frameworks engage with the production processes of AI systems, not only their downstream impacts.

The AI Dialogue can play a distinct role in advancing international cooperation by addressing a coordination problem in current AI governance: the uneven development and limited alignment of different governance domains. International efforts have made substantial progress in areas such as AI ethics, safety, and technical standards. However, other dimensions—particularly those relating to labour, production processes, and global value chains—remain less systematically integrated into these frameworks. This has resulted in a fragmented landscape, in which different aspects of AI systems are governed in relative isolation from one another. The Dialogue provides an opportunity to improve coherence across these domains. By convening governments, international organisations, researchers, industry, and civil society, it can facilitate a more integrated understanding of AI as a socio-technical system embedded in global economic structures. In practical terms, this could involve: Linking existing governance frameworks, including those related to human rights, labour standards, and AI governance Identifying areas of misalignment or gaps across these frameworks, particularly in relation to AI supply chains Supporting shared analytical baselines, including better understanding of how AI systems are produced and how value and risk are distributed Enabling iterative development of governance approaches, which may include both the adaptation of existing frameworks and, where necessary, the articulation of new principles or guidance. In this sense, the Dialogue's primary contribution lies in strengthening coherence and cumulative learning across governance efforts, while creating conditions under which more targeted normative or policy interventions can emerge.

Fairwork (www.fair.work), and their Director Prof Mark Graham (mark.graham@oii.ox.ac.uk)

Data workers involved in AI-related tasks remain insufficiently represented in global AI governance discussions. This reflects a combination of structural factors. Many such workers are engaged through intermediated or platform-based arrangements, often with limited access to collective representation. In addition, they are frequently located in regions that are geographically and institutionally distant from global policy processes. As a result, their experiences and perspectives are not consistently incorporated into discussions on AI governance, despite their direct role in the development of AI systems. Greater inclusion could be supported through: Structured opportunities for input from worker representatives, labour organisations, and civil society groups working on digital labour Targeted support for participation from stakeholders in regions where AI data work is concentrated Collaboration with intermediary organisations that can help document and convey worker experiences Mechanisms to ensure that such inputs inform ongoing processes, including summaries and follow-up work. Broadening participation in this way would contribute to a more empirically grounded understanding of AI systems and their governance.

https://fair.work/en/fw/principles/ai-principles/