From data annotation to reviewing disturbing material, human workers help build and maintain AI systems. Investigations across several countries have raised questions about pay, working conditions and accountability.
THE UNIVERSAL RECORD
Sourced reporting. No opinions.
Brad Socha | October 17, 2026 | 9:56 AM EST
Artificial Intelligence is often presented as technology that learns, reasons and produces results almost independently. Behind those capabilities, however, are people who label images, evaluate answers, transcribe recordings, assess safety risks and correct mistakes.
Much of this work happens far from the offices of the companies developing the systems.
Workers in countries including Kenya, India and the Philippines have participated in the data preparation and evaluation processes that support modern AI. Others work through online platforms, completing individual tasks for clients they may never meet.
Their contributions are essential to many AI development methods, yet the employment arrangements behind them can be difficult for the public to see.
The work also raises questions about how a rapidly expanding industry distributes its economic benefits, protects workers and accounts for the human effort behind increasingly automated services.
The Global Workforce Behind AI
Human involvement takes several forms.
Data annotators identify objects in photographs, categorize documents, verify speech transcripts and label information used to train or evaluate machine-learning systems. Other workers compare responses from competing AI models, identify factual errors or judge whether an answer follows instructions.
Specialists with backgrounds in mathematics, programming, science and other fields may assess more complex outputs. Some assignments are performed by employees; others are outsourced to contractors or distributed through digital labour platforms.
A widely reported example emerged in January 2023, when TIME published an investigation into work performed in Kenya for OpenAI through the outsourcing company Sama.
The investigation, based on interviews, payslips and contractual documents, found that Kenyan workers reviewing disturbing text for an AI safety project received approximately US$1.32 to $2 per hour in take-home pay, depending on their role and performance.
The documents indicated that OpenAI paid Sama a substantially higher hourly rate for the contracted service. That difference did not represent profit alone: the contractor also had employment and operating expenses. Nevertheless, it illustrated how payments from AI developers can be separated from workers’ actual earnings.
Workers interviewed by TIME described psychological distress associated with reviewing graphic descriptions of violence and sexual abuse. Sama said employees had access to professional mental-health support, while workers questioned the adequacy of the assistance.
The companies ended the project before its originally scheduled completion. The reported pay rates and conditions concern that historical contract, not a verified assessment of OpenAI’s current suppliers or compensation practices.
The investigation exposed a broader issue: AI systems designed to avoid harmful content can depend on people repeatedly examining precisely that material.
Not all annotation work involves disturbing content, and conditions vary considerably between employers, countries and assignments.
However, the International Labour Organization and independent labour researchers have documented concerns across digital platform work, including income uncertainty, limited bargaining power and the difficulty of enforcing employment protections across borders.
What Workers Are Paid, and Who Is Accountable
Compensation in the AI labour market is difficult to summarize with a single figure.
Some workers receive hourly wages. Others are paid per completed task, accepted annotation or project. Skilled evaluators may command considerably higher rates than workers completing basic labelling assignments.
Advertised rates do not always reflect actual earnings. Time spent qualifying for assignments, waiting for work, correcting rejected submissions or handling unpaid administrative tasks can reduce effective hourly compensation.
In its 2025 Cloudwork Ratings report, the Fairwork research initiative assessed 16 digital labour platforms against standards covering pay, conditions, contracts, management and worker representation.
Nine platforms scored no more than two out of ten. The findings concerned the platforms assessed and should not be interpreted as a measure of every AI data worker or employer. They nevertheless showed significant gaps in documented labour protections.
The structure of outsourcing complicates accountability.
An AI developer may commission work from a data-services company, which can distribute assignments through additional providers or platforms. Workers may have little visibility into the ultimate customer, while the customer may have limited direct contact with those performing the tasks.
Confidentiality agreements and commercially sensitive contracts can further restrict public scrutiny.
This creates practical questions about responsibility. Who determines whether the pay is adequate? Who provides psychological support when assignments involve traumatic material? Who investigates disputed payments or unfair removals from projects?
Partnership on AI, an organization involving technology companies and other stakeholders, has developed guidance recommending living wages, clear task instructions, channels for worker feedback and greater transparency across data supply chains.
Its April 2026 guidance also calls for stronger oversight of vendors and public reporting about data-enrichment practices. These are recommended standards, not universal legal requirements, and their adoption does not independently establish compliance.
There are potential benefits to the work. AI-related contracts can create employment opportunities, including remote positions and skilled technical assignments. Workers may gain income, experience and access to international clients.
But those benefits depend on the quality and stability of the jobs, not simply on the number of assignments offered.
The next stage of AI development may change the workforce further. Automated labelling and AI-generated training data could reduce demand for some routine tasks, while increasingly sophisticated systems may require more specialized human evaluation.
The balance remains uncertain.
What is already established is that modern artificial intelligence involves substantial human judgment. Workers help determine which examples systems learn from, which outputs are acceptable and which errors need correction.
As AI becomes more widely used in workplaces, public services and everyday products, scrutiny of its development is likely to extend beyond algorithms and computing infrastructure.
The conditions under which people train and evaluate these systems are part of the technology’s production process. Understanding who performs that work, how they are compensated and who is responsible for their welfare is essential to understanding the industry itself.
Sources:
TIME — Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic
https://time.com/6247678/openai-chatgpt-kenya-workers/
Fairwork — Cloudwork Ratings 2025
https://fair.work/wp-content/uploads/sites/17/2025/05/Fairwork-Cloudwork-Report-2025-FINAL.pdf
Fairwork — Cloudwork Platform Ratings
https://fairwork.oii.ox.ac.uk/en/ratings/cloudwork/
Partnership on AI — Responsible AI Starts with the Data Supply Chain (2026)
https://partnershiponai.org/responsible-ai-starts-with-the-data-supply-chain/
Partnership on AI — Data Enrichment Sourcing Guidelines
https://partnershiponai.org/wp-content/uploads/2022/11/data-enrichment-guidelines.pdf
Partnership on AI — Protecting AI’s Essential Workers: A Pathway to Responsible Data Enrichment Practices
https://partnershiponai.org/protecting-ais-essential-workers-a-pathway-to-responsible-data-enrichment-practices/
Partnership on AI — Protecting AI’s Essential Workers: Vendor Engagement Guidance and Transparency Template
https://partnershiponai.org/protecting-ais-essential-workers-introducing-our-vendor-engagement-guidance-transparency-template/
International Labour Organization — Generative AI and Jobs: A Refined Global Index of Occupational Exposure
https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure
About the Author
Brad Socha is the founder of The Universal Record, focused on sourced, factual global reporting. Coverage includes international news, geopolitics, technology, and major developments.