Who is really behind the Artificial Intelligence (AI) revolution? Of course, engineers and programmers - but also a far more hidden workforce outsourced to low-wage economies. From chatbots to image generators, AI systems rely on thousands of workers labelling data, filtering harmful content, and refining algorithms. These workers are often underpaid and endure harsh working conditions that has proved to contribute to serious mental distress. The EU’s Corporate Sustainability Due Diligence Directive (CSDDD) holds promise to address corporate accountability for human rights abuses in supply chains. But will it safeguard the hidden digital workforce that powers AI, or will these workers remain invisible and unprotected in the race for technological progress?
In the broader tech industry, Meta made headlines when reports revealed that outsourced content moderators in Kenya were paid as little as $1.30 per hour to sift through graphic and disturbing content, often at great psychological cost. Similarly, outsourced moderators for TikTok in countries such as Colombia and Tunisia have endured traumatic experiences due to the nature of the content they moderated. Now, the same patterns are emerging in the creation of Large Language Models (LLMs), where in one case, Kenyan workers outsourced by OpenAI were paid under $2 per hour to make ChatGPT safer by reviewing and labelling disturbing content, including suicides, murders and child abuse. Several affected workers have been diagnosed with severe post-traumatic stress disorder due to prolonged exposure to such material. In contrast, other industries that expose workers to traumatic content have recognized mental distress as an occupational hazard and introduced protections. For example, numerous U.S. states now allow workers’ compensation claims for PTSD in first responders even without a physical injury, acknowledging that mental trauma alone can be an occupational injury. Similarly, journalists covering traumatic events have increasingly gained access to structured mental health support. Yet, AI data annotators often lack comparable protections, leaving them vulnerable to long-term psychological harm. The mental impacts for the annotators are often worsened by job insecurity, low wages and poor working conditions, which are common in such precarious employment. These conditions create an environment where mental distress makes workers even more vulnerable to exploitation. This represents a fundamental human rights issue in not only the AI industry, but the tech industry as a whole.
The invisible workers powering AI
Data annotation, the process of manually labelling text, images, or video, is essential for making AI systems accurate and effective. To understand the nature of this work, data annotation can be compared to the simple tests we all complete to prove we are not robots - except the tasks these workers perform are far more complex, repetitive, and often distressing. Unlike typical remote tech jobs, the workers frequently review disturbing content or perform tedious labelling tasks under pressure. Yet, the people performing these tasks often remain invisible in discussions about AI development and its impacts. As the Partnership on AI, a coalition of AI organizations, acknowledges in a Time interview, 'Despite the foundational role played by these data enrichment professionals, a growing body of research reveals the precarious working conditions these workers face.' This invisibility may therefore not be accidental - AI companies often celebrate efficiency gains while obscuring their reliance on this hidden workforce. As the coalition puts it, 'Out of sight is also out of mind.' So, while tech companies refine AI’s ethical and environmental guidelines, they continue to overlook the fundamental rights of the workers whose labour is essential to the development of their product.
Much like traditional supply chains in industries like garment manufacturing, AI labour is often outsourced to low-wage economies where workers receive minimal pay, lack benefits, and face job insecurity. The data annotation workforce of AI represents yet another example of the power imbalance between corporations in the Global North and underpaid workers in the Global South, where valuable products are built on the backs of exploited workers. However, AI labour is still evolving - offering an opportunity to build ethical supply chains from the start and avoid the exploitative patterns of traditional industries.
Can the law keep up?
Despite growing awareness, companies and regulators have done little to address these challenges. Investigative journalists have exposed the exploitation of data annotators, forcing some companies to issue vague statements about addressing the issue. However, annotation workers are still being overlooked due to the fragmented nature of AI supply chains where work is outsourced across multiple jurisdictions and transnational accountability is unclear. These structures also make it difficult for workers to unionize or advocate for better conditions.
The lack of a unified legal framework further complicates efforts to ensure fair wages and safe working conditions for these workers. Directives like the CSDDD can therefore be fundamental in addressing digital labour and in ensuring that human rights’ protection is extended along the entire AI supply chain. The CSDDD mandates that companies conduct due diligence procedures on human rights and environmental risks across their entire supply chains.
Even though the directive does not explicitly mention digital labour, its broad provisions can apply wherever human rights risks exist. This presents both a challenge and an opportunity for the data annotation workforce: will regulators use the directive to hold AI companies accountable, or will these workers remain invisible? On the one hand, stronger due diligence requirements could push tech companies to provide fair wages and safer conditions for AI’s hidden workforce. This would be an important step towards fair treatment in the evolving digital supply chain. On the other hand, the lack of specific mention of digital labour creates ambiguity. Clearer guidance and oversight from national supervisory authorities will likely be essential to ensure AI workers are not left behind.
If AI is the future, then so is the workforce behind it. It is time to stop treating these workers as an afterthought and start holding AI companies accountable. Regulators must recognize that ethical AI governance is not just about data privacy and algorithmic fairness - it is also about ensuring that the people who build these systems are treated fairly. Unless digital supply chains are treated with the same scrutiny as physical ones, the CSDDD risks leaving behind a big group of workers. This could undermine the future of sustainable and ethical AI. ‘Out of sight’ should no longer be ‘out of mind’.
