Beneficiary-Based Responsibility for AI Training Labour: Lessons from Kenya’s Gig Economy and Beyond

Abstract

AI systems are built from large volumes of outsourced human labour, including data annotation, content moderation, and reinforcement learning feedback. Much of this work is performed in low-and middle-income countries under conditions of low pay, insecure contracts, psychological harm, and weak legal protection. Existing frameworks in labour law, business, human rights, and investment law struggle to address these harms because they rely on forms of legal or contractual proximity that outsourced AI training chains are designed to weaken. This paper argues that responsibility for AI training labour should be grounded not only in formal employment or contractual ties, but also in benefit and structural influence. Drawing on the Beneficiary-Pays Principle from climate justice, it develops a conditional framework under which obligations arise where harms are foreseeable, firms exercise structural influence over labour conditions, and benefits are systematic and durable. It then proposes a two-layer institutional response: extending due diligence to digital labour chains and establishing an AI training levy to support social protection and remedy for affected workers.