AI World

AI World - the forgotten faces of AI

An action-research program to render visible the conditions of data workers within AI value chains and ensure their effective legal protection.

Affaire｜Justice environnementale｜Justice sociale

9 September 2026

Share on socials

Behind artificial intelligence, human labor

Artificial intelligence is frequently presented as a largely automated technology, predicated on algorithms, models, and immense computational power.

However, behind AI systems lies a substantial amount of human labor.

Women and men produce, prepare, annotate, verify, and evaluate the data utilized to train these models. Others moderate the content published on social networks or assess the outputs generated by AI systems, thereby concurrently training the latter.

They are referred to as data workers.

Before an AI system can recognize an image, comprehend an instruction, or generate a response, vast amounts of data must inherently be prepared by humans.

The tasks of data workers notably include:

identifying objects within an image or annotating texts;
evaluating and correcting the responses generated by an AI model;
reviewing and classifying content published online;
identifying violent, hateful, or illicit content.

These operations contribute to training, enhancing, evaluating, and securing the digital platforms and AI systems we utilize daily.

Nevertheless, this labor remains largely invisible—as do the occupational hazards to which those who perform it are exposed.

From observation to legal action

Contemporary legal frameworks provide several instruments to assert jurisdiction over these value chains.

Within the purview of AI-WORLD, Intérêt à Agir and its partners primarily mobilize two legal avenues:

→ the corporate duty of vigilance (devoir de vigilance), to mitigate risks and address infringements upon fundamental rights, as well as the health and safety of data workers within AI value chains;

→ the right to personal data protection, to scrutinize the collection and processing of workers' data, particularly when deployed for surveillance or algorithmic management purposes.

Action 1: The Duty of Vigilance to hold ordering companies accountable

AI relies on a global value chain. The companies that purchase these services have obligations. The OECD guidelines are clear: the duty of vigilance follows the value, not the sector. We are initiating an action based on the French Duty of Vigilance Law ...

→ Discover the action

Action 2: Coming soon...

This upcoming action will focus on the General Data Protection Regulation (GDPR) to denounce the illicit collection and abusive use of workers' personal data (work pacing, geolocation, task history) as tools for algorithmic surveillance. Protection must no longer be limited to users alone, but must extend to the data workers who make AI possible.

A globalized labor supply chain

Data work is organized on a global scale. A corporation may design a service in one jurisdiction and outsource annotation, evaluation, or moderation to workers situated elsewhere, predominantly in Africa, Latin America, or Asia.

Two prevailing models dominate:

Business Process Outsourcing (BPO): these activities are delegated to specialized corporate entities that directly employ the workers.
Digital labor platforms: workers, frequently operating as independent contractors, remotely execute annotation, evaluation, or data preparation tasks.

This structural organization enables the mobilization of a globalized workforce. However, it concurrently fragments legal liability by multiplying the intermediaries between the workers and the corporate entities that design or commercialize digital services.

Extensively documented risks

Academic research, journalistic investigations, and field inquiries conducted among data workers reveal recurring occupational hazards, notwithstanding the diversity of jurisdictions, activities, and employment forms involved.

These risks are notably associated with:

employment precariousness and inadequate compensation levels;
stringent productivity targets and intensive surveillance systems;
a lack of transparency during the recruitment process and the distinct vulnerability of certain migrant workers;
impediments to the exercise of collective bargaining rights or the reporting of grievances;
potentially severe risks to mental health and psychological well-being.

Content moderation provides a particularly acute illustration. Certain workers are compelled to daily review images or videos depicting homicides, sexual violence, suicides, self-harm, or child abuse.

Repeated exposure to such material can inflict substantial psychological trauma. These hazards may be further exacerbated by the structural organization of the work itself: high processing quotas, continuous surveillance, employment insecurity, restricted autonomy, or the inadequacy of psychological support mechanisms.

Consequently, these risks are not solely inherent to the tasks performed. They are intrinsically dependent on the manner in which AI value chains are organized.

AI-WORLD: Making the labor behind AI visible

This is the fundamental objective of AI-WORLD, a program developed by Intérêt à Agir in collaboration with researchers, legal practitioners, trade unions, and worker advocacy organizations across Europe and Africa.

Our methodological approach integrates field research, social sciences, and strategic litigation to comprehensively understand working conditions within AI value chains and to identify the liability of the economic actors structuring them.

Because a responsible artificial intelligence cannot be assessed solely through the metrics of its models' performance or security.

It must equally be evaluated based on the working conditions of the individuals who facilitate its existence.

Our partners

DiPLab – Institut Polytechnique de Paris
Interdisciplinary research group specializing in digital labor, platforms, and artificial intelligence production chains.
Reversing.Works
Organization specializing in the reverse engineering and technical analysis of digital systems.
noyb – European Center for Digital Rights
European organization dedicated to the enforcement of digital rights and the protection of personal data.
CGT-FAPT
Trade union federation representing workers in the postal and telecommunications sectors.
Data Labelers Association – Kenya
Kenyan workers' organization striving for the recognition and defense of data workers' rights.
The Oversight Lab – Kenya
Nairobi-based organization leveraging research and strategic litigation to safeguard human rights within the technology sector.