Across the world, from factory floors to call centres, employees are increasingly asked, directly or indirectly, to hand over the raw material that will be used to train AI-powered robots designed to automate their own tasks. Others simply do their job as usual, without knowing that every word they say to a customer may later be used to train an AI model.
This article draws mainly on sources showing that Indian workers film their own gestures to train AI. This situation is not yet the standard, but it’s already drawing reactions from our journalists. Are we sawing off the branch we are sitting on by training AI to do our own work?
Employees end up training their own replacement
AI systems trained on text and images still struggle to reproduce simple human gestures. To solve this problem, companies now pay people to record themselves performing ordinary tasks (called “egocentric data”). Egocentric data means first-person footage meant to teach future AI-powered robots how to reproduce human gestures. For example, every call handled by a customer service agent becomes information an AI model can learn from.
In New Delhi, it’s common to see a saleswoman in a clothing shop wear a smartphone on her head for up to four hours a day while she folds clothes and prepares orders. For example, from an article of the media “Le Monde” (see source), her employer has partnered with Cogito Tech, a company that collects egocentric data.
But what’s the point for the employee in accepting this? One hour of footage is paid around six euros. The financial benefit works for both the employee and the employer. In the example from the article of “Le Monde” (see the sources) a shop owner openly admits that he would replace his staff with robots if they became cheaper than human labour. Nevertheless, this could also allow employers to use robots to ease the workload of human employees. For employees, even knowing they could lose their jobs in the future if robots became able to do the same tasks, the immediate financial need matters more in the moment. Moreover, sometimes the employee is not informed that the recording would be used to train AI or robots. The demand for this type of first-person video footage on employees is rising quickly because robots need very large volumes of images for their training.
This situation raises serious questions regarding legal issues. Are workers clearly informed that their image and gestures will be used to train a commercial AI model, or do they only realise it afterwards? If similar practices took place in Europe, they could struggle to comply with the GDPR, since data protection law requires clear information about the purpose of data collection, and freely given, informed consent, especially when biometric elements such as gestures or facial features are involved.
Consequence: Human financing his own replacement
Some of this work indirectly funds the automation that may later replace it. Companies collecting this data are honest about their goal: enough footage to build robots capable of replacing manual work entirely. Workers know this, because the extra income is immediate and the risk of automation feels distant.
In some sectors, companies openly present AI agents able to replace most of a customer service team. Other companies use AI tools to evaluate call quality, and later use these evaluations to justify layoffs. This creates an unusual situation. The same employee’s daily performance can be measured today, and used tomorrow as a reason for their dismissal.
Unlike a company investing in a machine it owns, here it is the employee’s own gestures, words and expertise that slowly become training material. This transfer of value is not something current employment law was designed to name, and even less to compensate.
Conclusion :
None of this means that recording is inherently wrong because the employee accepted it. The future challenge for employers could be: « Should employees have a right to know, or even a right to compensation, when their own work becomes training data for automation? »
More broadly, if the people best placed to teach a machine how to do a job are the very people who currently hold that job, what do we owe them once the machine has learned enough?
Source :
- https://www.lemonde.fr/pixels/article/2026/08/19/en-inde-le-paradoxe-des-travailleurs-qui-filment-leurs-gestes-du-quotidien-pour-entrainer-les-ia-je-suis-en-train-de-scier-la-branche-sur-laquelle-je-suis-assis_6749933_4408996.html
- https://www.aljazeera.com/gallery/2026/6/11/photos-indias-workers-are-training-ai-robots-to-take-their-jobs
- https://www.theguardian.com/global-development/2026/jun/24/indian-factory-workers-told-film-themselves-for-ai-robots
- https://futurism.com/artificial-intelligence/friends-colleagues-recording-conversation
- https://www.zonebourse.com/actualite-bourse/les-chatbots-ia-remplacent-les-travailleurs-des-centres-d-appels-en-inde-ce7d5adedd8dff2d
- https://www.businessinsider.com/ai-chatbot-ceo-laid-off-staff-human-support-2023-7
Étudiant en Master II Droit de l’économie numérique à l’Université de Strasbourg.
Centres d’intérêt : intelligence artificielle, protection des données, cybersécurité, histoire et droit des conflits armés.