Labour Rights

The people who make AI safe are paid two dollars an hour

OpenAI paid its outsourcing contractor roughly $12.50 an hour for each worker moderating the material that trained ChatGPT's safety filters. The workers received about $2. In February 2025 they founded an association.

P J Laszkowicz

On 30 October 2025, the Collaboration on International ICT Policy for East and Southern Africa convened DataFest Africa in Kampala.1 Among the sessions on AI governance and digital rights was a discussion most technology conferences never hold: the working conditions of the people who make artificial intelligence function.

Those people are data labellers and content moderators. They classify images, transcribe audio, evaluate model outputs, and review the most disturbing material the internet produces so that AI systems can learn what is acceptable and what is not. They work in Nairobi, Lagos, Bogotá, Manila, and Accra, employed by outsourcing firms whose names rarely appear in the product announcements of the companies they serve. They are paid between one and three dollars an hour.2 They are essential to every frontier AI model in production today. In late 2025, after years of documented exploitation, they began to organise.

What follows traces the labour supply chain beneath the AI industry, the harms it produces, the first organised responses, and the regulation, particularly the European Union's Platform Work Directive, that may start to change the terms. It is not an argument about whether AI should exist. It is an account of what it costs, and who pays.

The supply chain beneath the model

Every large language model, image classifier, content moderation system, and self-driving algorithm depends on human labour to work. That labour takes several forms: drawing boxes around objects and labelling them, sorting text by sentiment or toxicity, judging model outputs for accuracy and safety, transcribing and translating audio, and, most contentiously, reviewing and categorising the violent, sexual, and extremist material AI systems must learn to identify and filter.3

The work is overwhelmingly outsourced. The major AI companies contract with intermediary firms that recruit, manage, and pay the workers. The largest include Scale AI, which operates the Remotasks platform, Sama, formerly Samasource, along with Appen and CloudFactory.4 Those intermediaries recruit mainly in the Global South: Kenya, the Philippines, India, Venezuela, Colombia, Ghana, and Nigeria.5

The economics are stark. When OpenAI contracted with Sama for content moderation and data labelling to support ChatGPT, it paid the firm roughly $12.50 per hour for each worker.6 The workers received about $2.7 That is not an outlier. Hourly rates for data labellers in Kenya run from $1.50 to $2. In Venezuela the range is $0.90 to $2. Comparable work in the United States pays $10 to $25.8

The global data labelling market is projected to reach $1.89 billion in 2025, growing at nearly 24 per cent a year.9 The AI companies whose products depend on that labour are collectively valued in the trillions. The people performing it earn less per hour than a cup of coffee costs in the cities where those companies keep their headquarters.

Scroll. Click. Suffer.

On 28 May 2025 the human rights organisation Equidem published "Scroll. Click. Suffer", the most comprehensive investigation so far into the psychological harm done to data labellers and content moderators working for major AI and social media platforms.

The report draws on interviews with 113 workers across Colombia, Ghana, Kenya, and the Philippines, employed through outsourcing firms serving platforms including Meta, TikTok, and ChatGPT. A subset of 76 workers in Colombia, Ghana, and Kenya reported 60 separate incidents of psychological harm, among them post-traumatic stress disorder, paranoia, depression, anxiety, insomnia, and sexual dysfunction.10

The nature of the work explains the harm. Moderators review material including graphic violence, child sexual abuse, beheadings, torture, and sexual exploitation. Workers described handling 700 to 1,000 cases in a standard shift, with seven to twelve seconds to judge each one.11 The volume and the pace leave no room to process what has been seen. The content is not occasional. It is the whole of the job.

Equidem documented breaches of International Labour Organisation standards covering fair wages, the right to organise, and safeguards against forced labour.12 Where mental health support existed at all, workers described it as inadequate: brief counselling sessions offered by the same employer whose work caused the trauma, without independent clinical oversight or long-term follow-up.

The irony is structural. These workers perform what the industry calls safety work, the human labour that makes AI systems safe for everyone else. Making AI safe requires exposing people to precisely the material the system is being trained to detect and suppress. The industry's safety rests on their harm.

The TIME investigation and its aftermath

The crisis did not begin in 2025. In January 2023 TIME published an investigation by Billy Perrigo revealing that OpenAI had contracted with Sama to employ Kenyan workers reviewing graphic material to train ChatGPT's safety filters.13 They were paid less than $2 an hour. They were regularly exposed to depictions of sexual violence, child abuse, and murder. When they developed severe symptoms, the support offered fell short.14

The contract had begun in 2021.15 OpenAI acknowledged the work was challenging and described the workers' contributions as immensely valuable.16 When the contract ended, workers reported being sent home without pending pay or continued access to medical care.17

In 2023, 184 former Sama moderators sued the company over unfair termination and exploitative conditions.18 Separately, a group of Kenyan workers petitioned parliament to investigate the conditions they had worked under, asking the Ministry of Labour to examine both the outsourcing firms and the international technology companies they served.19

Sama was not the only case. In May 2022 Daniel Motaung, a South African recruited to moderate Zulu-language content for Facebook in Nairobi, sued both Sama and Meta.20 He had been paid about $2.20 an hour to review material including beheadings and child abuse. When he tried to organise a union to negotiate better conditions, he was dismissed.21

His case became a landmark. Meta's lawyers argued that the Kenyan court had no jurisdiction because the company was "not resident, trading or domiciled in Kenya".22 The court disagreed. On 20 September 2024 the Kenyan Court of Appeal ruled that Meta could be sued in Kenya, clearing the way for Motaung and more than 180 other moderators, who are seeking $1.6 billion in compensation, to take their case to trial.23

The case continues, and its significance runs past the individual claim. It established that the companies at the top of an outsourcing chain, and not only the intermediaries beneath them, can be held to account for conditions at the bottom.

Remotasks: a case study in abandonment

In early March 2024, workers in Kenya who logged into Remotasks, the data labelling platform run by Scale AI, found their access blocked without warning. Kenyan workers were cut off on 7 March; accounts were closed for good two days later.

Scale AI had ended access for workers in Kenya, Nigeria, and Pakistan at the same time.24 Workers in some countries received three weeks' notice. Those in Kenya did not, because of what the company later described as an internal error, attributing the shutdowns themselves to enhanced security protocols. Thousands of people who relied on the platform as their main income were left with no work, no notice, and no recourse.

A payment crisis compounded it. In December 2023 AirTM, the service through which Remotasks workers were paid, announced it was investigating money earned through the platform. Accounts were frozen, and AirTM later stated the funds had been improperly obtained and returned them to Remotasks.25 Workers who had done the labour and earned the money found their savings taken back by the payment intermediary, at the direction of the labour intermediary, with no adjudication and no appeal.

Kellion Mrego, on the platform since 2018, described staying with Remotasks through falling rates and thinning work, only to be removed without explanation or compensation.26 The pattern is characteristic: a workforce with no employment contract, no labour protections, no union, and no access to the legal systems of the jurisdictions where its employers are incorporated.

The Data Labelers Association

At the end of 2023 a group of Kenyan data workers formed a collective to press for their rights.27 They first tried to establish a trade union, and Kenya's labour framework makes formal registration difficult for workers classified as independent contractors on international platforms.28 In early 2025 they took another route. On 13 February, under the theme "Empowering the People Powering AI", they launched the Data Labelers Association.

Three hundred and thirty-nine workers joined in the first week. Within months membership passed 800.29

The Association's demands are not radical. It asks for fair wages, mental health support, transparent working conditions, and recognition of the work as skilled.30 It works alongside the African Content Moderators Union, the worker advocacy platform Turkopticon, and the Distributed AI Research Institute founded by Timnit Gebru after her departure from Google.31

It is already engaging Kenyan policymakers, talking with the Ministry of Information, Communications, and Technology to help legislators understand platform-mediated data work and to press for protections covering workers whose employers sit in foreign jurisdictions.32

The Association is the first organised labour response to data labelling conditions in East Africa. It is small, under-resourced, and operating in a legal environment not built for the work its members do. Its existence still marks a structural shift: the workers at the bottom of the AI supply chain have stopped waiting for the companies at the top to reform themselves.

From Nairobi to Brussels

At the other end of the chain a different kind of response is taking shape. Directive (EU) 2024/2831, the Platform Work Directive, was adopted by the Council on 14 October 2024, signed on 23 October, and published in the Official Journal on 11 November. It is the most significant piece of labour legislation to address AI-mediated work directly.

Its scope is wider than the name suggests. It covers not only ride-hailing and food delivery but any digital platform organising work through algorithmic management, explicitly including data annotation and content moderation.33 Member states have until 2 December 2026 to write it into national law.

Its provisions go to the mechanisms through which data workers are exploited.

It creates a legal presumption of employment for platform workers meeting specified criteria, shifting the burden from the worker proving they are an employee to the platform proving they are not.34 That directly challenges the independent contractor classification allowing outsourcing firms to withhold employment protections, benefits, and social security.

It bans algorithmic processing of certain personal data about workers, including their emotional or psychological state, their personal beliefs, and biometric data beyond authentication.35 For someone reviewing graphic material for eight hours a day, a prohibition on monitoring emotional state is not abstract. It protects them from systems that could penalise visible signs of distress.

It requires human oversight of significant decisions. A person doing platform work cannot be dismissed on the basis of a decision taken solely by an algorithm.36 Set against the documented pattern of workers removed from platforms without explanation, that establishes a minimum of procedural fairness.

And it guarantees workers information about how algorithmic systems are used to manage and assess them, along with the right to contest those decisions.37

The Directive is, in principle, the framework the Data Labelers Association is demanding through organising. The question is whether it reaches the workers who need it most.

The jurisdictional gap

The Platform Work Directive applies to platform work performed in the European Union. It does not apply in Kenya, the Philippines, Colombia, or Ghana, where most data labelling and content moderation is done.

This is the structural limit of even the most progressive labour law: it protects workers inside its jurisdiction while the work moves to jurisdictions without protection. If the Directive makes employing data labellers in the EU more expensive or more complex, companies can shift more of the work to countries where it does not apply.

There are forces pulling the other way. The Motaung ruling established that Kenyan courts can exercise jurisdiction over companies such as Meta even where they are not incorporated. The African Union's Continental AI Strategy, now in its first implementation phase, includes governance frameworks and capacity building for AI labour.38 Kenya's National AI Strategy carries a $1.14 billion allocation addressing the country's place in the global AI economy.39 Nigeria's Data Protection Commission introduced rules in September 2025 requiring impact assessments for automated decision-making.40

These frameworks are new. They lack enforcement capacity, institutional infrastructure, and in many cases the political will to challenge an outsourcing model that delivers cheap labour to the world's wealthiest companies. The distance between European legislation and conditions in Kenyan labelling centres is not a regulatory failure in any single country. It is a designed feature of a supply chain built to exploit the differences between legal systems.

The ILO's decent work framework, the OECD's AI principles on inclusive growth, and the UN Global Digital Compact all set out standards that would transform this work if applied.41 None binds anyone. None carries an enforcement mechanism. They are aspirations rather than obligations.

What fair AI labour looks like

The question is not whether the AI industry depends on human labour. It does. The question is whether that dependence will be acknowledged, paid for, and governed, or whether it will keep being placed in countries where neither journalists nor regulators reliably reach.

Fair AI labour would require several structural changes at minimum. Wages should reflect the difficulty, skill, and psychological toll of the work rather than the poverty of the market where it is performed. Workers should hold employment status rather than contractor classification, with the protections employment brings: social security, healthcare, paid leave, and protection from arbitrary dismissal. Mental health support should be independent, clinical, and continuing, not a short counselling session offered by the employer whose work caused the harm. Workers should be able to organise and bargain collectively without the dismissal Daniel Motaung and others have faced. And the companies whose products depend on this labour should carry direct responsibility for the conditions it is done under, rather than placing layers of intermediaries between themselves and it.

The Platform Work Directive addresses several of these, within the European Union. The Data Labelers Association is pressing for them in Kenya through organising. The Motaung litigation is testing whether they can be enforced through courts.

No one of these mechanisms is sufficient. Together they are the earliest stage of a response to a problem the AI industry created, benefits from, and has shown little appetite for solving unprompted.

The cost of intelligence

Every interaction with a frontier AI model rests on a foundation of human labour. Someone classified the training data. Someone judged the model's outputs. Someone reviewed the graphic material the model was taught to recognise and refuse. Someone was paid less than two dollars an hour to do it—and may have developed post-traumatic stress disorder in the process.

Total AI investment in 2025 is measured in hundreds of billions of dollars. The annual cost of fairly paying the data labelling workforce would be measured in the low single-digit billions, a fraction of one per cent of the capital being deployed. The disparity is not a technical limit. It is a choice.

The Data Labelers Association, the Equidem investigation, the Platform Work Directive, and the Motaung litigation do not individually change that arithmetic. They establish something that did not exist before: a record, a set of demands, a legal framework, and a precedent. The industry can no longer claim ignorance of the conditions under which its most essential labour is performed. What remains open is whether it answers that knowledge with structural change, or keeps placing the cost of its intelligence on the people least able to carry it.

Footnotes

  1. CIPESA, "CIPESA at DataFest Africa 2025: Advancing Ethical AI and Digital Rights in Africa," November 2025. ↩

  2. Wage ranges documented by TIME, Equidem, CBS News, and the LSE Media Blog across investigations from 2023 to 2025. ↩

  3. The taxonomy of data labelling work is described in the OECD AI Policy Observatory's reporting on AI labour and in academic literature on the "ghost work" supporting AI systems. ↩

  4. Scale AI operates Remotasks as its primary crowdsourcing platform. Sama, formerly Samasource, has been a primary contractor for OpenAI and Meta. Appen and CloudFactory are additional major intermediaries. ↩

  5. LSE Media Blog, "The perilous future of AI work in the Global South," 14 November 2025. ↩

  6. WeeTracker, "OpenAI Paid Sama $12 An Hour Per Worker—Kenyans Only Got $2," 25 November 2024. ↩

  7. TIME, "OpenAI Used Kenyan Workers on Less Than $2 Per Hour," January 2023. Workers reported earnings of $1.46 to $3.74 per hour, with most earning approximately $2. ↩

  8. The Conversation, "AI is a multi-billion dollar industry. It's underpinned by an invisible and exploited workforce," 2024. Venezuelan rates from the same source. ↩

  9. Market projection from industry analysis cited in o-mega, "Data Labeling Industry Guide," 2025. The compound annual growth rate given there is 23.6 per cent. ↩

  10. Equidem, "Scroll. Click. Suffer: The Hidden Human Cost of Content Moderation and Data Labelling," 28 May 2025. ↩

  11. Ibid. ↩

  12. Ibid. ↩

  13. TIME, "OpenAI Used Kenyan Workers on Less Than $2 Per Hour: Exclusive," January 2023. ↩

  14. Workers reported developing PTSD, paranoia, depression, anxiety, insomnia, and sexual dysfunction. TIME, January 2023. ↩

  15. OpenAI's contract with Sama for content moderation and data labelling began in 2021. ↩

  16. OpenAI statement quoted in TIME, January 2023. ↩

  17. Workers reported being sent home without pending compensation or continued medical care when the contract ended. ↩

  18. MedioNama, "Kenyan Workers Expose Disturbing Work Conditions in AI Data Labelling for OpenAI," July 2023. This action against Sama is distinct from the Meta litigation described below, and the number of claimants reported across the two cases varies between sources. ↩

  19. TechCrunch, "Workers that made ChatGPT less harmful ask lawmakers to stem alleged exploitation by Big Tech," 14 July 2023. ↩

  20. openDemocracy, "Daniel Motaung: Facebook slammed for trying to dodge Kenyan whistleblower," 2022. ↩

  21. Motaung was paid approximately $2.20 per hour and was dismissed after attempting to organise a union. openDemocracy and Africanews, 2022. ↩

  22. Meta's jurisdictional argument cited in TIME, "Meta's Attempt to Dodge Major Kenya Court Case Fails," 2023. ↩

  23. Meta Platforms, Inc & 2 others v Motaung & 186 others, Civil Appeal E232 & E445 of 2023 (Consolidated), KECA 1262 (KLR), 20 September 2024. Reporting on the case gives the number of co-claimants variously as 184, 185, and 186; the figure in the case name is used here for the citation and a rounded form in the text. ↩

  24. Rest of World, "Scale AI's Remotasks is booting workers with no explanation," March 2024. ↩

  25. AirTM announced in December 2023 that it was freezing worker accounts and returning funds to Remotasks on the grounds that they had been improperly obtained. ↩

  26. Mophat Okinyi, "Impact of Remotasks' Closure on Kenyan Workers," Data Workers' Inquiry, 2024. ↩

  27. Computer Weekly, "Kenyan AI workers form Data Labelers Association," February 2025. ↩

  28. Kenya's labour framework presents difficulties for workers classified as independent contractors on international platforms, making trade union registration hard to achieve. ↩

  29. Martijn Arets, "Data Labelers Association speaks up for invisible workers," 2025. ↩

  30. The Association's core demands are fair wages, mental health support, transparent working conditions, and recognition of data labelling as skilled work. ↩

  31. Computer Weekly, February 2025. ↩

  32. Computer Weekly, 2025; BMZ Digital.Global interview with Joan Kinyua, president of the Association. ↩

  33. Analysis by AI Policy Lab Sweden confirms explicit coverage of data annotators and content moderators. ↩

  34. Crowell & Moring, "New EU Directive Impacting Digital Platforms and Individuals Working for Them," 2024. ↩

  35. Article 7 of the Directive prohibits processing personal data on the emotional or psychological state of a person performing platform work, data drawn from private conversations, data used to predict trade union activity, and biometric data beyond authentication. ETUI analysis, 2024. ↩

  36. Fisher Phillips, "New EU Platform Work Directive Impacts Freelancers and Gig Economy," 2024. ↩

  37. Mason Hayes Curran, "The Platform Work Directive," 2024. ↩

  38. The African Union Continental AI Strategy, Phase I (2025–2026), focuses on governance frameworks and capacity building. ↩

  39. Kenya's National AI Strategy (2025–2030) includes a $1.14 billion allocation. ↩

  40. Nigeria's Data Protection Commission introduced the General Application and Implementation Directive in September 2025, requiring data protection impact assessments for automated decision-making. ↩

  41. The ILO Decent Work Agenda, the OECD AI Principles, particularly Principle 1.2 on inclusive growth, and the UN Global Digital Compact all articulate standards relevant to AI labour conditions. ↩