How Algorithmic Systems Govern Kenya's Content Moderators - AlgorithmWatch

How Algorithmic Systems Govern Kenya's Content Moderators

An exclusive survey of AI workers in Kenya reveals how automated management affects their livelihoods. Unions and advocacy groups are beginning to fight back.

Story

June 24, 2026

#ai #kenya

Lilian* (names followed by a star have been pseudonymized) is a former content moderator for a major US-based outsourcing firm in Nairobi. She describes repeated periods of being placed “on the bench” for months at a time, leaving her without stable income or certainty about when work would resume. For Lilian, the "bench" wasn't just a temporary pause, it was a financial death sentence. "It caused more suffering, [and put] financial constraints [on basics] like rent, food, and my livelihood." She says the job often demanded long hours and intense workloads, yet compensation remained extremely low and disconnected from the effort required.

Vivian*, who worked for the same company for over a year, describes a similar regime of metrics-driven exhaustion. "The system began measuring the number of tasks completed per hour more aggressively," Vivian explains. "Even minor drops in productivity or disagreements with automated quality assessments could affect bonuses, shift schedules, and overall job security." The unrealistic performance targets, stress, and anxiety defined much of her experience in digital labor. She recounts being placed on the bench and never being called back, a decision that left her struggling to meet basic needs such as rent, food, and daily survival.

Lawrence* describes a workplace transformed by stricter productivity quotas and increasingly aggressive automated performance tracking. According to his account, workers were expected to process more tasks per hour while simultaneously meeting higher accuracy thresholds, even when reviewing disturbing or emotionally taxing material. He says bonuses, schedules, and contract renewals were closely linked to system-generated scores, creating constant anxiety because minor declines in metrics could threaten both income and job security. In his view, the public rarely sees the human cost hidden behind efficiency targets.

These are the human faces of what researchers call the "regime of entrapment," a system where workers are lured by the promise of "jobs of the future", only to find themselves trapped in a cycle of high-stress, low-pay labor governed by opaque algorithms. In this industry, being "placed on the bench" means being suspended without pay while remaining "on call," effectively preventing workers from seeking other employment while they wait for a system that may never call them back.

Promises vs algorithmic reality

Algorithmic management refers to the deployment of automated systems to guide, assess, and oversee workers. While commonly associated with ride-hailing and delivery services, it has quietly become fundamental to outsourced digital labor. Business Process Outsourcing (BPO) firms, which contract for major technology companies, heavily rely on algorithmic management systems to supervise their employees.

According to recent confidential testimonies, gathered through a worker survey and direct outreach for this article, staff are typically evaluated by metrics such as tasks completed per hour, accuracy scores, error rates, decision speed, idle time, attendance, and quality assurance flags. Such metrics are then utilized to determine shift assignments, bonuses, contract renewals, and disciplinary actions. Effectively, workers are perpetually monitored and scored by software systems with little transparency or opportunity for appeal.

Recruiters present these roles as "jobs of the future," leveraging the aspirational appeal of Artificial Intelligence. Many workers enter the sector believing it offers economic advancement and a pathway into the technology industry. Recruiters frequently promote these positions as stepping stones toward AI careers. However, workers often discover a starkly different reality.

Trauma and the Tigray conflict

The human cost becomes sharper when algorithmic systems are applied to highly sensitive content. For Aman*, a content moderator originally from Ethiopia's Tigray region, the trauma was deeply personal. He recalled one of the most difficult periods occurring during the Tigray conflict, when he was required to assess large volumes of ambiguous and highly sensitive content, often depicting the very conflict impacting his homeland and people. "The system started flagging contents that were very ambiguous," Aman says. In a conflict zone, the line between hate speech and news reporting is often blurred, yet the automated system demanded instant, binary decisions. This ambiguity, coupled with the personal connection to the suffering depicted, made it harder to maintain the speed and accuracy demanded by employers. "This affected our reviewer competencies and bonuses," Aman adds, highlighting how algorithmic failure directly translates into lost income for workers. Beyond financial penalties, Aman emphasized that repeated exposure to violent conflict material can have lasting mental health consequences and may also create security fears for moderators and their families, especially when the content directly relates to their own communities.

According to that testimony, workers could be penalized financially for failing metrics even when the content itself was complex and unclear. This reflects a broader contradiction within content moderation: workers are expected to make rapid judgments on disturbing or politically sensitive material while automated systems continue to measure them as if all decisions carry equal difficulty.

Despite these severe challenges, Kenyan content moderators are beginning to establish organizations such as the Data Labelers Association (DLA) and the African Content Moderators Union (ACMU). These groups aim to advocate for fair compensation, transparent contracts, essential psychological support, and improved workplace safety measures. They contend that this exploitation is not an isolated incident but rather a systemic component of the global AI supply chain, emphasizing the urgent need for collective action and more stringent labor standards.

Systemic entrapment

The prospect of career progression in this sector is frequently misleading. The recruitment process itself is part of the trap. Intermediaries often target desperate individuals in Nairobi's slums, presenting these roles as prestigious tech jobs. However, the reality revealed through a confidential "worker survey" , distributed via encrypted Telegram and WhatsApp groups was far different.

"I can be fired any time of the day," says Andrew, a veteran of the industry. "And they will employ another person and pay them the same peanuts."

For workers like Lawrence, who has spent months on the bench, the message is clear: in the eyes of the system, they are as replaceable as the data they label. As the global AI industry grows, the voices of those building it remain scored, monitored, and ultimately, silenced.

Exiting the industry can be equally arduous. Many workers dedicate years to data labor without acquiring recognized certifications or transferable experience that could facilitate a move into other professions. For example, one accounting graduate recounted being confined to data work for eight years. Another worker encapsulated the problem by noting that despite maintaining numerous tailored CVs for online tasks, he effectively possessed “no CV.” Non-compete clauses and the lack of formal acknowledgment of their skills often keep many workers trapped within the AI supply chain.

Kenyan data workers are not calling for the elimination of these jobs, but rather for a fundamental re-evaluation of their treatment. Dinah, an experienced worker, critically posed the question, "Why do these laws function in their own countries? These tech companies are aware of what is right and wrong. Yet, they retreat to Africa to offload these jobs onto us, exploiting our desperation."