Earlier this month, footage of workers in an Indian garment factory went viral. Each worker sat at a sewing machine with a small camera mounted on their head (The Bridge Chronicle). No company confirmed the purpose of those cameras. No regulator launched an investigation (Indian Startup News), yet the clip prompted an uncomfortable question that researchers and trade unionists have been raising for years: are workers in the global south being recorded in order to train the automated systems that will eventually replace them?

The concern is grounded in how physical-labour AI systems are built. To replicate a manual task, a machine requires first-person footage of a human performing it, capturing hand movements, coordination and spatial judgement (WWMS). A head-mounted camera worn by a garment worker produces exactly this data at minimal cost. Whether that is what the footage depicts remains unconfirmed, but the scenario is technically plausible and consistent with documented industry trends.

Consent, data, and the absence of regulation

If garment workers' physical movements are being recorded to train AI systems, those recordings constitute biometric data. In most jurisdictions, collecting such data requires explicit informed consent. No evidence exists that the workers in the viral footage gave any (The Bridge Chronicle). Nor does Bangladesh, whose garment exports account for 80% of national export earnings, have any formal policy governing AI deployment in its factories (Global Information Society Watch). A 2026 peer-reviewed study of AI governance in Bangladesh's garment sector found that data protection frameworks remained largely unenforced at the factory level, and that workers had no meaningful avenue to understand or challenge how their data was used (Srivastava et al.). Union membership in the sector stands at between 5% and 10%, and unionisation is effectively banned in many facilities (Global Information Society Watch), making informed refusal practically impossible.

The distributional challenge

The World Economic Forum's “Future of Jobs” Report 2025 projects 92 million job displacements globally by 2030 alongside 170 million new roles created (World Economic Forum), a net positive that disguises a structural mismatch: 77% of new AI-related roles require postgraduate qualifications (ALM corp), while only 5% of Bangladeshi garment workers have access to any formal training (Global Information Society Watch). The WEF has warned of the emergence of an "AI precariat", a class of workers displaced not only from income but from the social identities employment provides, and has called for new policy mechanisms to address the psychological and civic consequences of mass occupational displacement (World Economic Forum).

What is required

Addressing these risks requires action on three fronts. Internationally, supply chain due diligence legislation, of the kind taking effect in Germany and France, must explicitly cover AI-related labour harms, not only wages and physical safety. At national level, India and Bangladesh need data protection regimes that apply meaningfully to factory settings, with enforceable consent requirements for biometric data collection. And at brand level, AI governance in supply chains must be subject to the same disclosure and accountability standards as carbon emissions or living wage compliance. As Christina Hajagos-Clausen of IndustriALL Global Union has stated: "If workers do not get to have a say about how it impacts them, they are at a disadvantage as a class." (Thomson Reuters Foundation)

The Rana Plaza collapse of 2013, which killed more than 1100 garment workers, demonstrated the human cost of regulatory failure and corporate impunity in global supply chains. Compensation cases remain unresolved more than a decade later (Amnesty International). The introduction of AI into those same supply chains, without transparency, consent or worker representation, risks repeating that failure on a larger scale.