Hidden Labour, Fractured Accountability: AI Data Annotation Supply Chains from Industry 4.0 to Industry 5.0

Industry 4.0 has advanced automation, data-driven optimisation and digitally integrated production, but it has also generated hidden social costs within the supply chains that make artificial intelligence possible. This article examines one such cost: data annotation labour. It argues that data annotation should not be understood as neutral preprocessing, but as constrained judgement labour through which human judgement, exposure and disciplined labour are converted into AI capability. Using Industry 5.0 as a normative lens, the article develops a conceptual and policy-oriented analysis based on secondary empirical synthesis and mechanism-based comparison. It examines Kenya, India and China as mechanism-revealing and maximum-variation contexts: Kenya illustrates outsourced harm handling, India illustrates quantified governance, and China illustrates state-embedded AI data production. Across these contexts, the article identifies a recurring condition of accountability fracture, defined as the structured separation of labour contribution, institutional authority, evidentiary visibility and remedial capacity across AI supply chains. The article shows how accountability fracture generates distributive, procedural and epistemic injustice, while undermining the human-centricity, sustainability and resilience promised by Industry 5.0. It then proposes a governance framework that reconnects control with responsibility, metrics with labour conditions, harm with remedy and workers with governance. The article concludes by arguing for a shift from human-in-the-loop to humans-in-the-governance-loop, in which data workers are recognised not merely as hidden inputs to AI systems, but as participants in the governance of the systems their labour makes possible.