The Blood and Sweat of AI: Millions of Hidden Workers Behind Every ChatGPT Query

Behind every answer from ChatGPT, Gemini or Claude stands an invisible army of people: hundreds of thousands of annotators and moderators in developing countries, without whom modern AI simply would not work. Their labor is carefully hidden.

Who is behind AI

Large language models do not learn on their own. They need people to label data, rate the quality of answers, remove toxic content and teach the model to tell a good answer from a bad one. Without this manual work, RLHF (reinforcement learning from human feedback, the key technique that makes models useful) is impossible.

This work is done by millions of outsourced workers through platforms such as Scale AI, Remotasks, Appen and Clickworker. Most are in Kenya, Pakistan, India, Venezuela, the Philippines and other countries where people can be hired for a few dollars an hour.

• Data labeling: identifying objects in images, categorizing text, assigning tags
• RLHF rating: comparing two model answers and picking the better one, explaining why
• Moderation: reviewing streams of generated content and deleting anything harmful
• Red-teaming: deliberately trying to break the model to find security vulnerabilities
• Transcription and translation: preparing training datasets for multilingual models

How much they earn

Rates range from $1 to $5 an hour, which is still little even in countries with a low cost of living. Most workers have no guarantees at all: no employment contract, no benefits, no protection against sudden termination of a contract. There are tasks today; tomorrow there are none.

Scale AI, valued at $13.8 billion, builds its business on exactly this workforce. CEO Alexandr Wang has openly talked about creating "the largest army of annotators in the world," yet the working conditions of that army are, as a matter of policy, absent from corporate press releases.

In 2023, Time journalists found that workers in Kenya hired through the contractor Sama to moderate content for OpenAI earned about $2 an hour. They spent hours reading descriptions of violence, which caused psychological trauma.

"We are the secret that companies prefer not to reveal," says one annotator from

Nairobi, who works for three different platforms at once.

Why it is hidden

AI companies sell a narrative of autonomous systems that learn by themselves. Acknowledging the enormous human labor behind every model breaks that image and raises uncomfortable questions about responsibility.

Legally, the arrangement is built so that no direct relationship exists between OpenAI and a specific annotator in Lagos; there is a multi-layered chain of contractors. This lets companies leave these workers out of official employment statistics and avoid obligations to comply with labor standards.

Researchers at the Oxford Internet Institute called this phenomenon ghost work. It is absolutely necessary for modern AI systems to function, invisible to the end user, and deliberately placed outside corporate responsibility.

What this means

The AI boom has created a new class of digital precariat around the world. While investors pour billions into autonomous systems, hundreds of thousands of people earning $2–3 an hour make those systems possible every day.

Without transparency in data supply chains, the tech sector risks reproducing in the digital economy the same problems long criticized in physical manufacturing, with one difference: factories can be seen, and data centers cannot.