Femnet e.V.

Digital Inequality - News and press releases

20 July 2026 Who pays for AI?

How much human work is actually behind AI tools such as chatbots, image generators, automatic content filters on social media, search systems, facial recognition or translation? Most of the time, this work remains invisible, and we haven't thought much about it yet. Precise figures do not exist, but estimates estimate that there are about 154 to 435 million data workers worldwide.

Millions of people who World Bank estimates Prepare, test, label or control data for AI systems. Their work often happens in the background, but is central to making artificial intelligence (AI) work at all, so that the tools can better recognize and process it. These include, for example, marking images, sorting texts, checking search results, moderating problematic content, and cleaning and sorting training data.

Exploitative business model, non-transparent conditions

Working conditions are often unsafe and stressful. Many work via platforms or subcontractors, often with little protection, short planning security and payment only per task instead of per working time. Low wages and mental stress are the rule and seem to be a distinct business model. The industry is strongly characterised by non-transparency and employees often have to sign so-called non-disclosure agreements (NDAs), which oblige them to remain silent about their concrete activities. In addition, there is algorithmic monitoring: The work is digitally controlled, evaluated and sometimes simply not paid for with poor evaluation.

The psychological burden arises not only from precarious working conditions, but above all from constant contact with disturbing content, lack of support and extreme insecurity. Especially in content moderation, people often have to see violence, hate speech or sexual violence, which can trigger anxiety, stress, sleep problems and, in severe cases, traumatic consequences.

Women* particularly stressed

Many of the data workers find themselves in vulnerable life situations where they have little access to better jobs. Women* are not only affected by the poor conditions in the industry, but they often have a stronger impact due to their gender-specific character. This mainly concerns a lack of security, psychological stress, unfair pay and a lack of professional recognition. In many cases, insecure employment relationships hit women particularly hard because they often have less bargaining power and less protection at work. This is combined with a double burden of care work that they have to do outside of their working hours. At the same time, they are more often confronted with sexualized or violent content during moderation and labeling activities, which can put a heavy psychological strain on them and reactivate their own experiences of violence.

Profitable exploitation of digital supply chains

Much of this work is systematically outsourced to the Global South. Companies in Europe or North America assign the tasks to platforms and service providers in countries such as Kenya, India, South Africa, the Philippines or Venezuela, because wages are lower there and liability is often more difficult to enforce. This creates a kind of responsibility gap: The companies benefit, but the risks and burdens end up with the local employees. This model is very profitable for Big Tech: By outsourcing, companies can buy large amounts of data work cheaply and thus build AI systems faster and cheaper. The profits are not only generated by software itself, but also by the exploitation of cheap, invisible work along the global AI value chain. Big Tech has a strong interest in hiding these precarious conditions and cracking down on government regulatory attempts.

What can we do?

Data workers make possible what we perceive as ‘artificial intelligence’. Without their work, many of today's AI tools, filters and automations would not exist in this form. This is why the topic is not only about technology, but also about fair pay, health protection and real corporate responsibility. So what can we do?

Individual AI users can …

 

Companies using AI tools can use …