THE SIGNAL IN ONE SENTENCE
On September 9, three organizations sent a formal notice to Teleperformance SE in France under the country's corporate duty-of-vigilance framework. The Data Labelers Association in Kenya, African technology and human-rights group The Oversight Lab, and French trade union federation CGT-FAPT say Teleperformance's parent company has not adequately identified or prevented risks facing people who moderate online content in Nairobi. Their notice is based on accounts from current and former workers. Those workers allege that some recruits learned the true nature of the job only after hiring, some migrant employees worked without required permits or faced delayed social-security contributions, and moderators were expected to assess disturbing material at a pace of roughly 45 to 60 seconds per item. They also describe sleep problems, exhaustion, depression, isolation and fear of retaliation. These are allegations, not findings by a judge, and the public materials do not disclose the number of workers interviewed or the social-media client. Teleperformance's own vigilance plan says moderation programs use its Welev8 wellbeing standards, on-site psychologists, paid wellbeing time, monthly counseling encouragement and an external assistance service. The worker accounts challenge whether those protections are accessible, trusted and effective in practice. The organizations gave the company three months to revise its plan and respond before they may seek action in the Paris Judicial Court. The plain signal is that a queue viewed in Nairobi can create a governance duty in Paris. The useful question is not whether a company owns a wellbeing policy. It is whether a worker can use it without losing time, pay, immigration status, privacy or a future contract.
01
WHAT ACTUALLY CHANGED
The Data Labelers Association, The Oversight Lab and CGT-FAPT issued the formal notice on September 9. Intérêt à Agir, a French public-interest legal organization, supported the action and published an English account. The organizations announced it during the International Network of Digital Labour conference at the International Labour Organization in Geneva.
The notice targets Teleperformance SE, the French parent of the TP group, rather than only the Nairobi operation. France's duty-of-vigilance framework requires covered companies to publish and implement a plan addressing serious human-rights, health, safety and environmental risks across controlled operations and certain business relationships. The signatories argue that Nairobi moderation risks belong in that parent-level plan.
The public account says the investigation involved current and former Teleperformance employees in Nairobi who moderated content for an unnamed social-media platform. It does not disclose how many workers participated, how they were selected, when each person worked, or which client contract governed the job. That limits any claim about prevalence while leaving the reported experiences important evidence to investigate.
Workers allege that recruitment materials did not clearly explain the true nature of content moderation and that some migrant employees worked for months or longer without required permits. They also describe delayed or missing social-security contributions and limits on freedom of movement linked to immigration paperwork. No reviewed public record establishes how many people experienced each problem.
The signatories say workers reported an average of 45 to 60 seconds to review an item while encountering violent and disturbing material. That time figure comes from worker accounts, not a published company metric or an independent time study. It nevertheless points to the central design problem: exposure, speed and performance management interact rather than arriving as separate workplace risks.
Reported health effects include sleep disorders, exhaustion, depressive episodes, addictive behavior, isolation, relationship strain and intrusive images that persisted after employment. Workers also said productivity targets could make it hard to reach psychological support and that confidentiality concerns reduced trust. The public record contains testimony, not a clinical prevalence study or an assessment of every Nairobi moderator.
The notice asks Teleperformance to map and rank data-worker risks, assess conditions with workers and unions, make recruitment transparent, protect migrant-worker and collective rights, reduce traumatic exposure, provide independent specialized psychological support during and after employment, improve grievance channels, work with clients on prevention and publish measures of effectiveness.
Teleperformance's vigilance plan presents a different account of its safeguards. It says the Welev8 program sets core wellbeing standards, each client program has a wellbeing coordinator and approved on-site psychologists, moderators receive a job description centered on moderation, paid wellbeing time is built into work, monthly individual counseling is encouraged and an external assistance program remains available for six months after an assignment. Those are company policy statements. They do not independently establish what every Nairobi worker was told, received or trusted.
02
WHY THIS MATTERS
Content moderation is the sanitation system of the social internet. Platforms depend on people to classify material that automated filters miss, misunderstand or escalate. The work can also help generate enforcement examples and operational knowledge used around automated systems. That does not make every moderator an AI worker, but it does place human judgment inside a technology supply chain that product stories often erase.
Moving the legal question to the French parent changes the scale of possible prevention. A local manager can adjust a shift. A parent company can set minimum recruitment, immigration, staffing, exposure, clinical-support, audit and client-contract standards across countries. It can also decide whether a contract that cannot fund those protections should exist at all.
The notice joins risks that corporate reports often separate. An unclear job description affects informed consent. A delayed work permit affects bargaining power. A speed target affects exposure and access to counseling. Confidentiality affects whether support is used. Contract renewal affects whether workers report a problem. A plan that counts each item alone can miss the machinery connecting them.
Psychological support is not proved by the existence of a counselor. Useful care needs protected time, genuine confidentiality, appropriate expertise, language and cultural fit, continuity after a worker leaves, and protection against retaliation. Utilization numbers also need care: low use can mean low need, or it can mean workers do not trust the door.
Clients shape the job even when another company signs the employment contract. Their content rules, queue volumes, accuracy targets, turnaround times, escalation design and pricing determine what an outsourcer can deliver. A serious remedy therefore needs contract-level changes and shared responsibility rather than a wellness workshop placed beside an unchanged production target.
Migrant status can turn an administrative failure into a coercive condition. A worker whose right to remain or move depends on employer paperwork may be less able to refuse an assignment, report harm or change jobs. Immigration compliance should be independently auditable and resolved before work begins, not treated as a back-office task that eventually catches up.
The three-month period matters because it creates a documented gate between notice and possible litigation. It gives Teleperformance time to explain existing controls, dispute allegations, propose changes or revise its plan. It does not guarantee mediation, admission, settlement or a court case. The next useful evidence will be the response and any measurable commitments attached to it.
France's framework is being used here as a supply-chain instrument rather than a border. If the approach survives scrutiny, workers and unions elsewhere may gain a route to ask parent companies for prevention data, not merely local promises. If it produces only broader policy language, the lesson will be less encouraging and still worth recording.
03
WHERE IT COULD HELP
- Give every applicant a plain-language preview of the content categories, likely intensity, performance measures, available support, client confidentiality limits, pay, contract term and right to decline before accepting the role
- Complete and independently verify work permits, social-security enrollment and related documents before a migrant employee enters production, with a worker-accessible record and remedy for delays
- Measure exposure by content type, severity, repetition and time, then use rotation, protected breaks, queue limits, escalation and stop rules instead of treating one average handling-time target as the whole job
- Provide confidential psychological support through specialists who are operationally independent from production management, with protected paid access, choice of provider and continuity after employment ends
- Separate clinical records from performance and contract-renewal decisions, publish who can access each data field and give workers a safe way to challenge improper use
- Add worker and union representatives to risk assessment, policy testing, incident review and client-contract renewal, then record where their recommendations were accepted or declined and why
- Require platform clients to disclose queue forecasts, high-severity categories, model and rule changes, quality targets and foreseeable workload spikes, and fund the staffing and care needed to manage them
- Report outcomes rather than amenities: permit completion, contribution timeliness, severe-content exposure, support wait times, protected hours used, grievances, retaliation claims, contract renewals, recovery follow-up and independently verified corrective actions
KEEP A HAND ON THE WHEEL
The notice records serious allegations, but it is not a judgment, regulatory finding or proof that every Teleperformance moderator had the same experience. The public release does not identify the social-media client, number of workers interviewed, interview protocol, employment periods, distribution of nationalities, complete evidence file or the full formal-notice document. The 45 to 60 second figure is worker-reported, not an independently measured average. Content moderation serves digital platforms broadly and should not be described as AI labor without evidence about the contract or data flow. Teleperformance's vigilance plan describes substantial wellbeing measures, but a policy description is not an independent audit of access, confidentiality, clinical quality or outcomes in Nairobi. No issue-specific company response was located before verification. Watch for Teleperformance's formal reply, a revised vigilance plan, named Nairobi measures, client participation, worker and union verification, immigration and social-security remediation, independently governed mental-health care, outcome data and any filing in the Paris Judicial Court after the three-month period.
04
TERMS WORTH KEEPING
SOURCES AND VERIFICATION STATUS
This article was written from the materials below. Product claims and dates were checked against those sources on September 19, 2026.
PUBLICATION RECEIPT: Revision 1. Published September 19, 2026.
THE PUBLICATION ENGINE
WANT A SIGNAL OF YOUR OWN?
We build source-grounded publications, private briefings, and editorial systems for organizations with something useful to say.
WORK WITH US