Across Nairobi and other Kenyan cities, a growing number of young people are quietly contributing to the development of Chinese artificial intelligence systems. Students and recent graduates, facing limited job opportunities, are being recruited to perform large scale data labeling work that trains AI models. This labor takes place largely out of public view, organized through informal digital networks rather than established corporate structures.

In the early hours of the morning, many of these workers sit in dark rooms reviewing thousands of short video clips. Their task is to identify details such as movement speed or specific actions, repeating the process again and again to meet strict daily targets. Communication with supervisors and teammates happens almost entirely through WhatsApp groups, which function as virtual factory floors where productivity statistics, rankings, and instructions are shared in real time.

Most workers earn about 700 Kenyan shillings per day, equivalent to just over five dollars, after working up to 12 hours. Payments are made through mobile money services such as M-Pesa. There are no written contracts, formal onboarding documents, or access to human resources departments. Instead, workers are added to projects after filling out basic online forms, often recommended by friends or classmates already in the system.

Kenya has long been a hub for outsourced data labor used by major United States technology firms. Companies such as Meta, Google, and OpenAI have relied on Kenyan workers through established outsourcing firms, arrangements that have previously drawn criticism over low wages and poor working conditions. In recent months, reporting shows that Chinese AI companies have begun tapping into the same labor pool, but through far less transparent methods.

According to workers interviewed, Chinese projects are typically managed by local middlemen or third party agents. Annotators rarely know the names of the companies behind the work or how the labeled data will ultimately be used. Some team leaders report brief video calls with Chinese managers, but most instructions are delivered automatically through messaging apps. Specialized platforms used for the work are often inaccessible to the public and provide little information beyond a login screen.

The pressure to perform is intense. Teams must maintain high accuracy rates, usually above 85 percent, to receive payment. During trial periods, entire groups risk being removed from projects if accuracy drops. Experienced workers learn to speed up by using multiple devices at once and relying on pattern recognition, a process some describe as mentally exhausting.

Kenya’s high youth unemployment rate has made this type of work appealing despite its challenges. As of July 2025, unemployment among young people stood at 67 percent, according to the Federation of Kenya Employers. The country’s strong English proficiency, reliable internet infrastructure, and favorable time zone make it attractive for global outsourcing, including AI data annotation.

Labor advocates and unions have raised concerns that these arrangements represent a new form of digital exploitation. They point to the lack of contracts, unclear accountability, and absence of legal protections for workers. Kenyan government officials acknowledge the issue and say current labor laws do not adequately cover this type of digital work. Authorities are now consulting with labor and technology ministries to develop regulations aimed at identifying responsible employers and safeguarding vulnerable workers.

As artificial intelligence continues to expand globally, the experience of Kenyan data labelers highlights the human labor behind automated systems. While the work provides income in a difficult job market, it also exposes gaps in oversight that policymakers are now under pressure to address.