University

To think about the implications and create regulations that ensure AI technologies are regulated and ethically developed.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Open-source software, open data and open AI models

There are myriad implications of AI, especially on the social and ethical scale. It is necessary to address these challenges from a human rights perspective because societies are currently grappling with fears primarily related to safety and equality. Women and children are already disproportionately affected. Workers within the AI supply chain are also adversely affected by having to oversee and moderate content and data that is fed into systems. Lastly, beyond human rights as we know them, climate change and polarisations have led to further complications. Therefore, whilst many might think that the technical regulation of technologies may suffice, in this case, AI deserves to be explored from the angle of human rights because its very existence hinges on invisible human labour.

The impact of technology on the development of law, and the extent to which current laws and regulations both capture and fail to protect human rights, is overlooked because of their tight overlap. It becomes necessary to understand how best to navigate intersectional aspects of legislation when dealing with ubiquitous technologies. The thematic areas highlighted above cover a range of issues; however, there is also a need to develop and implement new laws and regulations to help us govern societies in the digital age.

The rapid growth of AI development has led to a vast yet largely unseen labour infrastructure—a global network of data workers who label, annotate, and moderate content that trains and sustains AI systems. This workforce occupies a critical gap in governance: AI regulation often fails to address upstream labour harms, and human rights due diligence frameworks that could offer protection are simultaneously weakening. While much attention is directed at downstream harms from AI use, less focus has been given to human rights violations and environmental damages stemming from the upstream processes and materials essential for AI models. This represents a major governance gap: current AI regulations—such as the EU AI Act—primarily target AI outputs and deployment risks, with little regard for labour conditions during development. The core issue is structural: AI regulation focuses on what AI systems do, not on the effort required to create them. Human rights due diligence frameworks, which generally oversee supply chains, are declining even as AI supply chains grow. Consequently, labour exploitation, union suppression, and environmental harm along these supply chains remain inadequately addressed.

All thoughtful discussions can contribute to awareness building. What is predictable is preventable and effective measures can be accordingly discussed and implemented over time. Moreover, it is important for multiple stakeholders to be present and contribute to these discussions to ensure the newer and upcoming regulations remain inclusive and well considered.

Increased research opportunities. We are at a stage where the impact of these upcoming technologies is still unknown. To collaborate on understanding the long-term implications of AI, we need adequate research. Additionally, building a community where we can consistently discuss challenges and opportunities will truly matter for AI governance.

Governments and intergovernmental bodies should provide the institutional scaffolding: setting agendas, ensuring continuity between dialogue cycles, and translating consensus into binding or soft-law instruments. They are uniquely positioned to bridge the AI Dialogue with existing human rights mechanisms, including UN treaty bodies and the ILO. Civil society organisations and labour advocates are essential for surfacing harms that technical and corporate stakeholders systematically underreport — particularly those affecting workers in the AI supply chain, marginalised communities, and people in the Global South. Their participation must be resourced, not merely invited: civil society groups without funding or access cannot engage meaningfully with technically dense processes. Affected workers and communities — including data annotators, content moderators, and gig-economy AI workers — should have direct representation, not only through NGOs. Mechanisms such as structured testimony, worker delegate seats, and anonymous submission processes can help ensure the voices of those who face retaliation risks are heard. Industry and developers bring technical expertise and operational knowledge, but their participation should be balanced by conflict-of-interest safeguards and transparency requirements around lobbying positions. Researchers and academics can provide independent evidence, comparative analysis, and horizon-scanning, but should be protected from capture by either state or corporate funders.

Data and platform workers—such as annotators, content moderators, and gig workers—who support AI systems are almost entirely excluded from governance discussions, even though they face some of the most immediate harms. Their exclusion is fueled by insecure jobs, geographic dispersion, and efforts by intermediary platforms to suppress collective organising. Communities in the Global South play a significant role in AI's labour infrastructure but benefit little economically and have little say in the rules governing it. Although Africa, Latin America, and South/Southeast Asia make up the majority of the world's population, they have minimal representation in AI governance.

Implementing structured red-teaming and adversarial formats, where civil society actors are explicitly responsible for challenging industry and government positions, would add constructive tension to what might otherwise become exercises in consensus-building. Facilitating asynchronous and distributed participation—such as through written submission platforms with mandatory responses, translated materials, and regionally hosted satellite dialogues—would expand access beyond those able to attend in-person events in Geneva or New York. Maintaining living documentation—a shared, publicly editable record of dialogue outputs that updates in real time—would boost transparency and enable broader publics to monitor how contributions are reflected in outcomes.

Supply chain accountability frameworks. The Partnership on AI's Responsible Data Enrichment Guidelines recommend that AI companies implement internal governance, oversee impacts across the data enrichment process, and issue transparency reports on their supply chains. These guidelines are now part of the OECD's upcoming Due Diligence Guidance on Trustworthy AI, indicating a potential for global adoption. The DeepMind-PAI partnership provides a practical internal model for applying these principles. Worker-led organising as governance. The Kenyan Data Labellers Association demonstrates that workers are already independently organising, representing a bottom-up form of governance that should be supported, not ignored, by formal institutions. Mandatory human rights impact assessments-currently used in development finance and some national frameworks-should be expanded to evaluate AI systems and their supply chains prior to deployment, rather than after the fact.