Since 2022, Kenyan courts have become the forum where the question of who answers for the human labour behind commercial AI is being argued. Four years in, there is still no ruling on the merits, and no verifiable source documents higher wages or more stable contracts in Kenya’s data annotation sector. What changed is where the fight happens, not the hourly rate.
The job looks like this: a queue of content scrolling on a screen, a task counter, a key to assign a label. Sometimes the label is a category — violence, nudity, hate speech — sometimes it is a preference between two machine-generated answers. Either way, what the worker produces does not stay on that screen. It becomes training data. It feeds a moderation classifier or a large language model, and it teaches a machine to recognise, or to avoid, exactly what a human being has just looked at.
This is where artificial intelligence actually enters the Nairobi story, and it is not a side detail. Automated moderation systems and large generative models do not arrive clean from the research lab: they become usable through a phase of human supervision, in which people label toxic material so the model learns to filter it. A 2023 TIME investigation documented payments between $1.32 and $2 per hour to Kenyan microworkers employed to label toxic content for OpenAI. That is the point where the AI supply chain becomes visible — not in a data centre, but in an office in Nairobi.
Four years of litigation, zero rulings on the merits
On 10 May 2022, Daniel Motaung, a South African former content moderator, sued Meta and Samasource Kenya EPZ — Sama, the vendor that ran Facebook moderation for East Africa — in Nairobi, alleging forced labour, human trafficking, union busting and pay discrimination. Motaung stated he was paid around $2.20 an hour; the February 2022 TIME investigation that opened the case, Inside Facebook’s African Sweatshop, reported wages as low as $1.50 an hour. Sama had dismissed him in 2019, after he led more than a hundred colleagues in an attempt to unionise; the company maintains the dismissal was for bullying and coercion.
March 2023 opened a second front. A group of moderators — 43 at the outset, later a figure sources place somewhere between 184 and 186, with no single official number — sued Meta and Sama over unlawful mass dismissal, tied to the moderation contract moving from Sama to the Luxembourg-based firm Majorel. The court issued an interim order suspending the dismissals. In April 2023, the Employment and Labour Relations Court, with Judge Mathews Nduma Nderi presiding, asserted Kenyan jurisdiction and confirmed the interim order. In June 2023, a Kenyan court ordered Meta to provide moderators with “proper medical, psychiatric and psychological care”. In October 2023, mediation between the moderators, Meta and the contractors collapsed without agreement.
The heaviest moment came on 20 September 2024, when Nairobi’s Court of Appeal rejected Meta’s jurisdictional objection, sending both proceedings to trial. In April 2025, a separate case seeking around $2 billion over hate content linked to the conflict in Ethiopia was also found admissible before Kenyan judges. Then, on 12 February 2026, the awaited rulings in the two proceedings were deferred: ruling on notice, no date fixed.
It is worth stating plainly, because this is the part most retellings skip: nobody has won anything. What has been decided so far concerns jurisdiction and interim relief. No liability has been established. Meta has consistently argued it was never the moderators’ employer; Sama has called the allegations inaccurate.
The missing number, and why it is missing
The most common framing of this story — “after the lawsuit, what changed in wages and contracts” — assumes something that does not exist. There is no verifiable source documenting a measurable change in pay or contractual structure in Kenya’s data annotation sector attributable to those proceedings. TIME reported in 2022 that Sama moderators would receive a raise following its investigation, but the size, scope and duration of that raise are not verified — and in any case it would predate the lawsuit rather than follow from it.
A July 2025 report from Nairobi described the case of Naftali Wambalo, a data labeler working for Sama, earning roughly a dollar and a half an hour, and mentioned Sama and CloudFactory alongside the Kenyan government’s encouragement of outsourcing firms setting up shop. Three years after the investigation that started all of this, the order of magnitude is unchanged.
A warning about a trap in circulation: that report came from a Hong Kong outlet and the original figure was in Hong Kong dollars. Twelve Hong Kong dollars an hour is about one and a half US dollars. Anyone republishing the number without converting it tells the opposite of the true story.
What actually changed
Something did move, but along a different axis. Two things, precisely.
- Jurisdiction. Before 2024 it was far from obvious that a Kenyan worker could bring an American multinational that never formally employed him before a Nairobi judge. The Court of Appeal decision of September 2024 made that claim triable. The precedent touches the entire architecture of outsourcing: the separation between client and vendor stops being an automatic shield.
- Organisation. The African Content Moderators Union was founded by around 150 African AI workers, with recognisable figures such as Richard Mathenge, Kauna Malgwi and James Oyange. It is the first pan-African structure that treats moderators and annotators as a single occupational category rather than as appendages of separate companies.
Neither of these shows up on a pay slip. Both change the balance of power in which a pay slip will eventually be negotiated.
Why wages are the wrong metric to watch on their own
There is a structural reason why litigation does not translate into raises. Annotation work is engineered to be mobile. A moderation contract can move from one vendor to another — which is exactly what happened in 2023, when the work shifted from Sama to Majorel, triggering the second lawsuit. When the client is not the employer, a case won against vendor A does not bind vendor B, and vendor B can sit in another country. Legal pressure, unless it fastens onto the client, slides away with the contract.
That is why the jurisdictional fight matters more than it appears, and why the companies contested it so stubbornly. It was never only about Motaung or the dismissed moderators: it was about whether the place where the work is done is also the place where you answer for how it is done.
Kenya as a hub: a trend, not a policy
Nairobi is now routinely described as Africa’s epicentre for AI outsourcing: widespread English, low labour costs, government encouragement for firms to set up operations. But be careful about turning a trend into an industrial plan. No legislative act, official strategy or verifiable public figure formalising that candidacy is on the record. Nor is there a verified law on digital workers or data annotation. There is a market that formed, and litigation that grew alongside it. That is not the same thing as public policy.
The most useful thing to say to anyone following this from outside is also the least spectacular. The missing number — what a Nairobi annotator working on data destined for a commercial model earns today, on what contract, for how many hours — is not missing because nobody looked. It is missing because no public register of these wages exists, and the figures circulating on anonymous salary-reporting sites for Nairobi rest on a handful of mutually inconsistent entries. It closes only with primary sources: Employment and Labour Relations Court filings, documentation from the law firms and organisations supporting the claimants, pay slips shown by the workers themselves. Until that material is public, anyone writing that “wages went up after the lawsuit” is publishing a wish, not a finding.