Executive summary

This report exposes the institutionalised harm within the global digital labour system, which sustains the online experience for 5.2 billion people, or 63.9% of the global population. Content moderators review user-generated content to ensure platform safety, while data labellers provide context for machine learning models, such as OpenAI’s ChatGPT. Equidem engaged with 116 data workers across Colombia, Kenya, and the Philippines, including those subcontracted by Meta, ByteDance, and OpenAI.

Methodology

Research was conducted between August and October 2024, involving interviews with 113 workers. This sample included 42 individuals from business process outsourcing (BPO) firms subcontracted by Meta, 17 by OpenAI, and 24 by ByteDance. To protect participants from retaliation, all names used in this report are pseudonyms.

Part 1: Data workers in the digital platform economy

The digital economy is defined by unequal power relationships where lead firms hold disproportionate bargaining power in a buyer-driven market known as monopsony. This allows them to set terms that drive accelerated work for extended hours. Operations are concentrated in hubs such as Bogotá, Nairobi, and Manila, where BPO and knowledge process outsourcing services contribute significantly to national economies, accounting for 3.5% of the GDP in Colombia.

Part 2: Informality and exploitation of data workers

To meet high-speed turnaround times at reduced costs, downstream companies frequently hire workers informally through short-term contracts or without contracts entirely. A case study of Remotasks operations in Kenya illustrates how workers lack social protection and job security, often facing abrupt dismissals and forced unpaid overtime to meet undisclosed client targets.

Part 3: Target pressure and industrial discipline practices

Moderators are evaluated on productivity and accuracy, often facing unmanageable targets. Workers in Colombia and Kenya reported having only 7 to 12 seconds to moderate each case, reviewing 700 to 1,000 cases per shift. Discipline practices include pay cuts and the withholding of bonuses, which can comprise up to 70% of total remuneration. One worker in Ghana reported that his bonus accounted for 72% of his total income, leaving the base salary insufficient to meet financial needs.

Part 4: Health and safety risks and workplace violence

Sustained exposure to graphic content results in severe physical and psychological harms, including insomnia, hair loss, eye strain, and significant weight loss; one individual lost 15 kilos in 6 to 8 months. Psychological harms include PTSD, anxiety, depression, and suicidal ideation. Exposure to violent sexual content has caused sexual trauma and detachment. Mental health support is often ineffective, with some supervisors threatening workers who took well-being breaks exceeding 1.5 hours per week.

Part 5: Barriers to accountability

Structural practices of secrecy, including non-disclosure agreements (NDAs), prevent workers from reporting abuse or seeking care. Over 100 former moderators in Kenya filed a lawsuit against Meta and Sama for mental health issues and union-busting. Other legal actions include a Spanish court recognising a moderator’s trauma as a work-related injury and an investigation by Colombia’s Ministry of Labour into reports of traumatic conditions at Teleperformance.

Conclusion and recommendations

The report calls for a structural shift in governing digital labour. Immediate measures include lead firms upholding freedom of association and suspending gag clauses. Short-to medium-term reforms suggest the International Labour Organisation should recognise data work as high-risk essential labour. Performance bonuses should be capped at no more than 25% of total compensation to avoid creating perverse incentives. Long-term changes involve redesigning platforms to reduce human exposure to toxic material and establishing a global recovery and reparations fund for affected workers.