Worker Voice and Post-Deployment Governance in AI Supply Chains: Evidence from Unionised Workplaces

The supply chains of AI systems have been well-documented to create new forms of precarious labour. Our research shows that even traditionally privileged workers such as unionised workers in the UK are being co-opted into AI supply chains and are experiencing a corresponding erosion in their working conditions.  This paper examines such workplace experiences as part of a broader AI supply chain, and examines how workers and their representatives encounter failures, negotiate harms and attempt to influence governance of AI. 

Drawing on evidence from an interactive webinar with UK union representatives (n=422), practitioner workshops (n=65) and focus group discussions (n=14), the paper analyses how representatives experience and navigate AI after deployment. The workshops and focus groups also develop and examine the Worker-centred Reflexive AI Tool (WRAIT), a structured method for helping representatives document, articulate and negotiate AI-related concerns. Guided by an ethics-in-practice approach, which treats everyday experiences of uncertainty, unease, adaptation, and judgement as forms of ethical knowledge, the study uses these experiences as evidence of AI harms. Union representatives provide a valuable perspective because they sit between workers' lived experiences and organisational decision-making, making visible how concerns travel, or fail to travel, into governance processes. 

The findings show that worker influence over AI is constrained by fragmented and underdeveloped governance infrastructures. Representatives described unclear system ownership, limited consultation, weak escalation pathways and reactive, compliance-driven approaches to accountability. These conditions create an articulation bottleneck: workers and representatives are expected to identify and challenge AI harms, yet their experiential knowledge is often discounted in favour of technical, legal or managerial expertise. 

Representatives conceptualised AI risks more broadly than many corporate and regulatory frameworks allow. Their concerns extended beyond job security, pay and working conditions to include service quality, client wellbeing, professional standards, environmental impacts and long-term occupational futures, showing how risks are distributed across workers, organisations and communities. 

The analysis of WRAIT shows that structured reflection can strengthen worker voice and make governance gaps visible, but it cannot replace institutional mechanisms that embed social dialogue, accountability, and collective bargaining. 

By linking post-deployment workplace experience to debates on human-centric AI supply chains, the paper argues that responsible AI governance requires more than technical evaluation or procedural compliance. It depends on recognising workers' experiential knowledge and creating mechanisms through which that knowledge can shape how AI systems are deployed and governed.