One arm of the Assembly line of AI. EU data workers and privacy regulation

Data work (Muldoon, 2024) for AI training purposes is now a planetary market (Graham and Ferrari, 2022).

"AI producers tap into pools of disadvantages" (Casilli et al. 2025) for each country, reproducing historical colonial relations where Global South countries are exploited.

The AI WORLd research project however looks at the EU as another place for data work with one case study, on a well-known "ai-training" platform, advertised as more "high-end", and also aiming for a European and US workforce. Through our research (36 interviews, 94 data annotation documents, 11 audits of the website's behaviour shared by data workers), we evidence several elements.

Following the footsteps of Tubaro et al. (2020) and Miceli et al. (2024), the socio-economical status of these EU workers contradicts the platform's statements : while it sells an "expert" annotating workforce, young and educated workforce, the pay and work structure of the platform remain similar to other work platforms that don't advertise themselves as high-end, with a significant gap between these workers' real income and their needs.

We find significant evidence of personal data theft and surveillance through a wealth of trackers (16), including social media platforms. All 6 privacy risks that Sannon et al. (2022) mention are present, with two significant consequences.

One: this data should be considered extra labour, on top of the surplus vale that is extracted traditionnally through labour activities in the marxian tradition.

Two: these activities are certainly in breach of the European General Data Protection Regulation. 

As such, this indicates that, as much as Madagascar is a hub for data work in the AI production chain of French AI companies, the EU is becoming, or has become, a hub for data work and data extraction in the AI production chain of US companies.

The position of the EU in this international division of labour is thus the following : personal data is extracted from European citizens by U.S. companies, and then sent back for annotation to precarious E.U. data workers whose data is also stolen. Those services are then deployed in the European Union with a complete disregard for personal data regulation.

This research thus provides evidence for a strong data protection regulation in the interest of data workers, as well as institutions and organisations interested in digital sovereignty.

The last step of this project is strategic litigation : we offer support for EU workers to file complaints with EU data protection authorities. Our methodology is replicable in non-EU jurisdictions, which strengthens the case for defining personal data regulation also as work protection (see Colclough, 2022, and Molé, 2025). 

Data workers, if they can become active stakeholders in the processing of their own data, may also offer a better outcome for the general public whose data is collected for AI training through an alliance between worker and consumer unions (Le Bonniec, 2024).