Guardian Reveals Shrinking Tech Work and Uncertain Pay for Refugees in Kenya

Refugees in Kenya’s Kakuma camp are finding that remote tech work can mean completing assignments without knowing whether they will be paid. In an investigation published October 8, The Guardian interviewed more than two dozen refugees, documenting discretionary rewards and shrinking entry-level opportunities as AI tools spread through the work once promoted as a route to income.

A route into work narrows

Bahana Hydrogene established the Solidarity Initiative for Refugees’ digital-skills programs in 2016, hoping outsourcing would provide dependable income. The organization has trained more than 2,000 refugees and connected more than half with remote tech work, from short gigs to long-term employment.

Staff once readily found annotation, transcription and translation assignments. Hydrogene said workers typically earned 500 to 1,000 Kenyan shillings, or $3 to $6, a day before Remotasks, a Scale AI subsidiary, left Kenya in 2024.

The International Trade Centre estimates that transcription, data-entry, translation and web-research jobs have declined by about 50% since 2022, spokesperson Susanna Pak told The Guardian. That is an estimate of disappearing opportunities, not a measurement of how many jobs AI alone eliminated.

Research now, payment uncertain

Grace, a refugee who requested anonymity because of a nondisclosure agreement, used ChatGPT to research hymn translations in languages she does not speak. She submitted information through AOP Connect, a crowdsourced research platform owned by British software company RWS.

Materials reviewed by The Guardian said her work could earn up to $500 as a discretionary reward, rather than guaranteed wages. The project advertised $9,000 in available rewards, shared among an unknown number of contributors. SIR mentor Asha Luka Abdallah said most earn much less than the maximum without being told why.

In its response, AOP Connect said reward terms are communicated transparently upfront and payments reflect research quality, relevance and timeliness. It also said the platform is not AI-enabled or used for AI-model training. Grace’s use of ChatGPT does not make her assignment an AI-training job.

The language of ‘rewards’ instead of wages shifts risk entirely on to workers, who invest their time and expertise without any guarantee of fair compensation

Mark Graham, professor of internet geography at the Oxford Internet Institute, speaking to The Guardian

The harder climb into specialist work

Even workers using AI face pricing pressure. Hubert Sanga, founder of refugee-training enterprise Generation Aid, said assignments that once took days now take hours, cutting billable time. Training groups are pressing companies for higher pay and better conditions, while workers fear replacement by AI or cheaper labor.

Na’amal, which connects refugees with employers, is moving away from microwork—small tasks distributed through online platforms—and toward training in software engineering, digital marketing and e-commerce. Founder Lorraine Charles said clients increasingly want specialists rather than workers with basic digital literacy.

That shift has limits. Konexio director Fabien de Castilla said disrupted schooling and work experience make competing for highly skilled jobs difficult for most refugees. He sees possible future openings in data cleaning and reviewing AI outputs, but those remain potential opportunities, not a replacement income stream.