Behind The Screen: What Our 2026 Labor Conditions Report Found

Every image an AI model recognizes, every offensive post it filters, and every sentence a chatbot answers safely was shaped, at some point, by a human being sitting at a screen — drawing boxes, flagging content, correcting labels, one task at a time. That work almost never makes it into the story of AI. This year, the DLA set out to change that by asking the workers themselves what the job actually looks like.

Over the past several months, our team conducted structured interviews with data labelers and content moderators across Kenya, documenting pay, working hours, mental health support, and job security. What we heard was consistent enough to call a pattern, and serious enough to call urgent.

The gap between "AI work" and how it's treated

Data labeling is routinely described as entry-level or temporary — a stepping stone rather than a career. In practice, it demands sustained focus, judgment calls on ambiguous or disturbing material, and fluency with tools and guidelines that shift from client to client. The workers we spoke to are, in effect, the quality assurance layer for some of the most widely used AI products in the world — yet many described contracts that could end without notice, and take-home pay that had not kept pace with the platforms' growth.

"You are told you're replaceable right up until the moment your labels are the reason the model works at all."

What moderators are carrying

Content moderation surfaced a distinct set of concerns. Reviewing graphic or violent material for hours at a time takes a measurable psychological toll, and most workers said mental health resources — where they existed at all — were treated as optional rather than a basic condition of the job. This is one of the reasons DLA built mental health support into our core mission areas, not as an add-on benefit but as a standard we're asking employers to meet.

Where the wins are already showing up

It isn't only a story of gaps. Collective bargaining conversations this year contributed to measurable wage improvements for members in our network, and a growing number of platform partners have agreed to clearer contract terms and defined review-load limits. Change is slow, but it is happening — and it is happening because workers organized and were willing to speak on the record about conditions that are usually kept quiet.

What we're asking for next

The full findings feed directly into DLA's ongoing policy advocacy: transparent contracts with defined notice periods, wages benchmarked to the value the work creates rather than the cheapest available labor market, and mental health support built into every moderation contract, not offered as a courtesy. We're also pushing platforms to publicly disclose how much of their AI pipeline depends on human labeling and moderation — visibility is the first step toward accountability.

If you label data, moderate content, or manage a team that does, we want to hear from you. Every interview adds weight to the case we're building, and every member who joins makes the next negotiation stronger.