Billed as the ‘world’s first comprehensive AI law’, the EU AI Act (AIA) sets a precedent for the regulation of Artificial Intelligence (AI) – one that could influence policy-making globally. Referencing the whole ‘AI value chain’, AIA employs a risk-based legislative approach. Yet, prima facie, this constitutes an internal contradiction of the regulation. Given that risk is a fundamentally forward-looking concept predicated on the ‘unrealised potentialities’ of future events, any risk-based regulation offers a limited scope for regulating retrospective harms along the value chain. Certainly, by drawing on the theory of ‘AI empire’, this thesis explores how the AI value chain is saturated by the exploitation of ‘AI labour’ – that is, human labour which produces metadata for AI systems. In determining to what extent AIA addresses the exploitation of human-performed ’AI labour’ – my research problem – I undertake a qualitative content analysis of AIA using both data-driven and theory-driven coding. I conclude that AIA contains a wholly inadequate address of AI labour since AIA 1) contains no conceptualisation of AI labour; 2) fails to enshrine AI labourers’ rights by only protecting rights deemed ‘at risk’ by functional AIs; and 3) legislates solely against prospective risks late in the AI value chain.