Large technology corporations attempt to externalise the social costs of artificial intelligence (AI) development, including the hidden human labour. Data work tasks – including data cleaning, image labelling, transcribing, content moderation and, most recently, red teaming – are indispensable to continually train AI models and enhance the accuracy of their outcomes. These tasks are performed by a globally dispersed workforce who encounter decent work deficits such as exploitative wages, workplace surveillance, automated evaluation, absence of welfare benefits, poor mental health outcomes, arbitrary termination, and refusal of wages. Moreover, these workers are predominantly concentrated in the Global South even as they drive AI expansion in the Global North, demonstrating the inequities in global AI value chains.

Yet, data workers have remained invisible in public and policy discourse, despite numerous governance directives at national and global levels. The G20 Generic Framework for Mapping Global Value Chains is a useful starting point for investigating vulnerabilities in global value chains. This brief aims to build on this mandate to identify how workers’ rights are imperilled by fragmented and precarious AI value chains, and provide an evidentiary basis for policymaking in this regard. It argues for the development of a due diligence framework in AI value chains that ensures corporations and AI developers are accountable, and data workers’ human and labour rights are protected – a necessary step towards ensuring sustainability and inclusiveness in AI value chains.