Interviews with 10 Kenyan AI annotators show how Chinese companies hire data labelers via opaque middleman networks and WhatsApp groups to avoid accountability
Rest of World : LinkedIn: Dámiláre Dòsùnmú . Bluesky: @fireantprincess and @restofworld.org LinkedIn: Dámiláre Dòsùnmú : my latest story is about how chinese AI companies are quietly tapping into kenya's young workforce, hiring students and recent graduates … Bluesky: @fireantprincess : we already live in a cyberpunk dystopia restofworld.org/2025/kenya-c... [embedded post] @restofworld.org : Kenya's unemployment crisis has allowed Chinese AI companies to tap into a workforce of young people, who spend 12-hour shifts labeling data in thousands of videos for around $6 a day restofworld.org/2025/kenya-c...
Context & Ripple Effects
The reporting extends a documented pattern in which Chinese data-labeling vendors used vocational-school internship pipelines to obtain low-cost annotation labor. Here, informal intermediaries and WhatsApp recruiting appear to shift that sourcing into Kenya while further obscuring the employer-worker relationship.
Kenya was already part of AI’s outsourced data-work chain through Nairobi-based annotation operations and Kenyan labeling work used to improve ChatGPT. The new detail is the reported use of opaque networks by Chinese companies rather than a clearly identifiable direct contractor.
First-order effects
- Kenyan students and recent graduates recruited through these channels bear the immediate costs: reported 12-hour shifts, low daily pay, and limited clarity over who ultimately directs the work.
- Chinese AI companies can obtain annotation capacity while putting middlemen between themselves and the labor conditions attached to their training-data supply chain.
Second-order effects
- Clients, auditors, and labor advocates have less ability to trace working conditions when recruitment and payment are routed through informal brokers rather than named vendors.
- More transparent annotation providers may face price pressure from networks able to source labor with fewer visible contractual obligations.
Third-order effects
- If this model spreads, AI training-data work could become more fragmented and harder to govern, with accountability separated from the companies benefiting from the labeled data.
- The contrast with direct Kenya-based contracting suggests that labor standards in AI supply chains may increasingly hinge on whether buyers require traceable subcontracting, rather than on the location of workers alone.
The trend: AI’s expanding demand for human training data is driving labor-cost arbitrage through increasingly layered, cross-border contracting networks.