On Saturday, 17 October, the conference Hacker la Tech, organised by the French investigative media outlet Mediapart, took place in Paris. Bringing together professionals, researchers, writers, journalists, and activists, the event offered a space to discuss technology from a range of critical perspectives. One of the panels, introduced by Antonio Casilli and Dan Israel, a journalist at Mediapart, invited Caroline Wambui, trade unionist and co-founder of the Data Labelers Association, and Germán Salazar, a member of Corporación Ngen. The Data Labelers Association is a trade union founded by Kenyan data workers, advocating for the fair treatment and recognition of data labelers. They recently formally notified Teleperformance, along with Intérêt à Agir, The Oversight Lab Africa, and DiPLab, over its inaction in protecting workers’ integrity. Corporación Ngen is a Chilean collective founded by residents of Quilicura, a neighbourhood of Santiago where tech companies, notably Google, have settled their data centres because of the wetlands surrounding the city. These wetlands are currently threatened by data centres that use water to cool their servers. The discussion brought into focus two of the most important concerns surrounding the development of artificial intelligence: AI water consumption and environmental destruction, and the labour exploitation it produces. By bringing together perspectives from Latin America and Africa, the speakers highlighted the connections between struggles that are often perceived as separate and explored the “double cost of AI”. Whether these struggles concern the environmental consequences of data-centre infrastructure or the working conditions of those who produce the data, these movements confront different dimensions of the same technological model. The conversation revolved around the need to build collective forms of resistance, initially at the local level but, in the longer run, through collective action that can transcend geographical boundaries as AI issus are themeselves, globalized. Rather than treating environmental protection and workers’ rights as distinct causes, the conversation invited participants to consider how these struggles might converge in challenging the social and ecological costs of AI. Watch the video in french bellow

DiPLab’s report showing how Outlier tracks the workers that train Big Tech’s AI models was released yesterday, and the media have already started picking up on the story. This is how the story was covered in Spain, France, Norway, Poland and Denmark so far. Klassekampen (Norway) : “– American companies like Meta, Open Al and XAI are using Scale AI‘s platforms to exploit young, newly educated Europeans as a workforce to train their AI models” mediapart.fr (France) : “according to the conclusions of a report released on October 16, Outlier, a subsidiary of the American company Scale AI, itself a pioneer of the annotation since 2016, subjects its workers to extreme surveillance, harvesting and exploiting a very large number of their personal data.” Version2.dk (Denmark) : “In order to be able to work on the platform at all, the workers are forced to provide sensitive biometric data and facial scans through an invasive third-party app called Persona, which is owned by Peter Thiels Founders Fund.” Wyborcza.pl (Poland) adds : “a market has quietly emerged in which Outlier exploits highly skilled Europeans for not particularly high wages while simultaneously building a system of constant surveillance and data extraction that can be used for both AI training and commercial or military purposes” El Pais (Spain) concludes : “The authors of the report believe that Outlier is in breach of European regulation as it is today. They believe that the deregulation promoted by the Commission could weaken GDPR articles that function today as firewalls that stop the transfer of data and knowledge to the United States.”

DiPLab is releasing a report on the surveillance of data workers today. Entitled “Surveillance at Scale : How AI training platforms gather personal data from European workers” this report is a research project in collaboration with Reversing.works and Intérêt à Agir, funded by the Digital Freedom Fund and the European Artificial Intelligence & Society Fund. International technology companies collect vast amounts of personal data from European citizens. Public debate and policymaking often focus on data taken from consumers through e-commerce, social media, or streaming platforms. Yet many Europeans also use digital platforms to work and train AI systems. Online freelance platforms are usually discussed in terms of low pay and insecure “gigs”. Much less attention is given to another resource these platforms extract from workers: their personal data. This report examines the case of Outlier, a key data work platform for artificial intelligence. Outlier is operated by Scale AI, a company that recruits workers to perform tasks such as evaluating AI-generated responses, creating training data, and improving AI models. This report shows how digital platforms rely on data workers worldwide, including in the EU, to label, check, and organize training data. These data are harvested online. But Outlier also collects workers’ personal data, financial records, biometric information, and even facial recognition scans. This data is used for surveillance purposes, and probably for further AI training purposes. Our interdisciplinary team analyzes the evidence collected from qualitative interviews, network logs, and guidelines, to provide evidence of a large-scale data collection process, through hidden trackers. We also provide a legal analysis in terms of likely European law (GDPR) data protection violations : the AI WORLd project reveals how the EU has become a periphery for data extractivism in the production process of AI models. Ethical AI requires the ethical treatment of the modern-day factory workers whose work makes it possible. Beyond the economic exploitation data workers through inadequate compensation, this report reveals the extraction of personal data for purposes ranging from covert AI training to potential military applications. These practices raise serious concerns about social justice, labor law, and European data security. Surveillance_at_scale_diplab_october_2026Download To site this report : Thomas Le Bonniec , Antonio A. Casilli, Mathilde Abel, Simone Robutti, Gaetano Priori, Donatella D., Paul S. Surveillance at Scale : How AI Training Platforms Gather Personal Data from European Workers DiPLab / Reversing.Works report, 2026.

Utopia and Work: When Alternatives Become Possible, a collective volume edited by Mariagrazia Cairo Crocco, Enrico Donaggio, and Nadine Richez-Battesti, was published by Theseo on September 30. In his chapter, “Utopies concrètes et utopies nécessaires à l’heure des intelligences artificielles” (“Concrete utopias and necessary utopias in the age of artificial intelligence”), Antonio Casilli develops four possible utopias for data work, drawing on forms of resistance and agentivity already emerging among workers. Despite the harsh working conditions faced by data workers, the intense surveillance to which they are subjected, and the strict prescription of their work through algorithmic management, data workers engage in both individual and collective forms of resistance to platforms. Workers are resorting to different means of resistance to resist platform exploitation. These include strikes by drivers and delivery workers across Europe, the United States, and Asia, as well as legal action initiated by data workers in Eastern Africa. Through these different forms of action, workers have secured new rights and protections, including the reclassification of employment contract, access to social protections in cases of illness or work related accidents, or even minimal hours payment. Resistance can also be found in free labour done by platform and social medias users with boycott, consumers strikes and consumers unions. As platform workers and users increasingly repurpose platform architectures as tools of resistance, a key question emerges: How do workers invent new forms of autonomy and resistance beyond the constraints imposed by platform capitalism? This chapter builds on one decade of research conducted by Antonio Casilli and DiPLab on data work across the world. It proposes four concrete utopias, possibilities for alternative futures grounded in existing practices of resistance and collective struggle, complemented by two necessary utopias to counterbalance the limitations of the first four. These four concrete utopias are: (1) unionization, whether through data workers joining existing trade unions or creating their own; (2) platform cooperativism and the development of international alternative networks platforms; (3) legislative regulation and the adoption of rules governing platform work; and (4) accountability on the part of technology companies and platforms, through partnerships with their clients and regulatory bodies. The first of the two necessaries utopia is a Social digital income : aka the collective distribution of the money created trough users activities on platforms. The goal is to recognize the free labour users are providing to platforms by using them and also to ensure data workers are remunerated correctly The second is the re-territorialization of digital labour, which involves the collective reappropriation and local governance of digital platforms and infrastructures, with the aim of strengthening workers’ digital sovereignty and their control over the technologies that shape their working conditions. You can download the book for free here

In a recent article, entitled “Séances de psy, crises ésotériques, jurons… Les étranges expériences d’Anthropic sur la «conscience de l’IA» ” Elise Vianacourt asks Antonio Casilli what he thinks about the potential emergence of conscious AI. The article discusses whether AI could be conscious or not, and, if so, whether we should care about its well-being. This is already actually the case at Anthropic, which has hired a team to “take care” of its model. In his interview, Antonio Casilli reminds us that behind AI ethics research are the very same companies that produce AI themselves. This allows them to direct research in the direction they want, avoiding the issue of human work and the working conditions behind AI. “A major part of the research on artificial intelligence is financed by the big companies in the sector themselves, and that’s a problem. To put it another way, it gives them a way to choose their own research topics and create a fake opposition.“ Casilli also argues that this could be a way for tech companies to avoid any legal repercussions for violations of the law that their models could commit. Nothing more than a “tactical and marketing” move. “Imagine a scenario where, twenty years from now, we consider AI not as a galaxy of products, but as sentient beings with rights. Picture what it could look like in a court of law. Confronted with illegal behavior by its machines, Anthropic could argue that the company isn’t responsible. Even worse, it could say: “If you punish us, you are putting the well-being of this creation at risk, and you are violating its rights.” “ You can read the article here

On September 29th, Antonio Casilli was invited to the podcast Les Temps Électriques by Sophie Sontag-Koenig, alongside Arnaud Casado, a legal researcher and member of the Sorbonne’s Judicial Institute, who specializes in labour law. In this episode, entitled “Algorithms at Work! Sociological Investigations and Legal Mobilizations,” the two researchers question the transformations of labour caused by AI, the importance of CSEs (Social and Economic Committees), data workers’ rights, and the environmental toll of AI. Casilli’s interview opens with the “invisible work” carried out by data workers, far removed from the smooth image of the data scientists and engineers we imagine behind AI. He describes this vast category of workers who “teach” AI, label data, and moderate content, more often than not in countries of the Global South. He explains how tangible their work actually is and adds: “This work is far from invisible as soon as we start looking for the workers behind it.” Arnaud Casado considers the possibility of fighting against AI through the law, and especially through the relationship of subordination between workers and companies. One possibility could be the Duty of Vigilance Law. Recently, Teleperformance was formally notified to undertake a comprehensive review of its vigilance plan, by the Kenyan workers’ association Oversight Lab Africa and the Data Labellers Association. Casilli reminds us that Kenyan workers face very low wages, are denied labour rights such as maternity leave, and are often relocated by Teleperformance itself, which hires them from other African countries, leading to visa blackmailing, and illegal employment. This is the first time the Duty of Vigilance Law has been mobilized in this way. Later in the interview, Casilli links data workers and the environment by pointing out that data worker hubs and AI extractivist infrastructures, such as lithium mines and data centers whose voracious water consumption is endangering local communities, can often be found in the same places, mainly in the Global South. He then discusses the Chilean situation, where he is currently conducting research, and the case of activists from the Atacama Desert who are currently fighting against the development of data centers in the oasis they have inhabited for generations. He describes these areas in the Global South as “sacrificial areas”, where “human resources are sacrificed as much as environmental resources.” Casilli closes the podcast by reminding us to question the technologies that make our lives easier at work, or that are presented as “more ethical”. For example, Zoom is often presented as an alternative to in-person meetings, which are considered more polluting. Yet Zoom itself involves AI and therefore relies on the invisible, poorly paid work of thousands of workers across the globe, as well as on data centers and other polluting technologies. You can listen to the podast here

Mathilde Abel was recently interviewed by Aurore Maubian for Libération. The interview is available in French here or below. In this interview, she explains the “enshittification” of platforms and how ChatGPT is gradually testing and rolling out an advertising model in order to maintain a free product while generating revenue. She also highlights the consequences on individual health and well being of platforms increasingly intense competition for users attention. She also draws attention to the strong extractivist dynamics (ressources, data, work) underneath these business models. Pop-ups qui surgissent, publicités qui envahissent l’écran, options autrefois gratuites désormais payantes : sur les plateformes numériques, l’expérience des utilisateurs se dégrade immanquablement à mesure qu’elles gagnent en popularité. Un phénomène que l’écrivain et journaliste canado-britannique Cory Doctorow a vulgarisé, en 2022, sous un terme qui ne s’embarrasse pas de nuances : la «merdification», de l’anglais enshittification. Après les réseaux sociaux ou encore les plateformes de streaming, ChatGPT en offre un nouvel exemple. Depuis lundi 24 août, des publicités sont intégrées en France et dans une trentaine de pays européens, dans la version gratuite et l’abonnement le moins onéreux. Auprès de Libération, Mathilde Abel, socioéconomiste spécialiste des enjeux numériques et maîtresse de conférences à Télécom Paris évoque «un mécanisme connu depuis vingt ans sur le plan économique. On retrouve la musique de “si c’est gratuit, c’est vous le produit”.» Pourquoi OpenAI introduit-elle de la publicité sur ChatGPT en Europe ? OpenAI est une entreprise qui doit entrer dans une logique de rentabilité, puisqu’elle continue d’enregistrer d’importantes pertes financières. Avec la publicité, elle ouvre un nouveau segment de marché et ajoute donc une nouvelle source de revenus. Elle continue de proposer un service gratuit, mais celui-ci est désormais financé en partie par la consommation de contenus publicitaires. C’est un modèle économique très traditionnel et surtout très rémunérateur. Ce n’est pas pour rien que Google l’utilise sur son moteur de recherche ou sur YouTube, et que Netflix s’y est également mis. En conséquence, cela produit de fait une dégradation de la qualité de service pour les consommateurs qui vont continuer à utiliser ChatGPT gratuitement. Fallait-il s’attendre à ce que ChatGPT suive cette logique de «merdification» ? D’un point de vue stratégique, ce n’est pas surprenant qu’OpenAI introduise finalement de la publicité. L’entreprise suit la logique classique des modèles économiques des moteurs de recherche et de certains médias. Le cycle de vie de ces plateformes est bien connu des économistes. Il y a d’abord une phase de lancement, avec une recherche de liquidités qui permettent de subventionner le marché et donc l’usage dans un premier temps. C’est pour cela que vous, moi, tout un chacun, avons pu utiliser ChatGPT gratuitement pendant plusieurs années. -------------------------------------------------------------------------------- Ensuite vient la phase de croissance, pendant laquelle les plateformes multiplient les offres. ChatGPT l’a fait en conservant un service gratuit, mais aussi en proposant un abonnement premium, avec différents niveaux d’accès en fonction de ce que vous êtes en mesure de payer [8 euros pour l’abonnement Go, 23 euros pour le Plus et 103 euros pour le Pro, ndlr]. C’est ce qu’on appelle de la «discrimination tarifaire» : petit à petit, un monde se crée entre ceux qui ont la capacité d’accéder à des services premium et ceux qui ne l’ont pas. Puis vient la phase de «stabilisation», où les plateformes cherchent à maintenir leurs sources de revenus. Comment ces plateformes, malgré la détérioration de leurs services, parviennent-elles à rester attractives aux yeux des utilisateurs ? Avant de lancer ce projet, OpenAI a réduit cette incertitude quasiment à zéro. En amont, l’entreprise a testé le marché de manière très régulière pour savoir si, oui ou non, les consommateurs utilisant gratuitement ChatGPT resteraient sur la plateforme en présence de publicité. Cela se fait à travers de petits segments de marché : les plateformes ne mettent pas immédiatement de la publicité massivement partout. Elles peuvent proposer des publicités à certains individus et regarder comment ils réagissent à ce modèle. -------------------------------------------------------------------------------- Leur grande force est que nous produisons, comme usagers, énormément de données, ici notamment à travers les prompts [les requêtes envoyées à l’IA]. Ceux-ci leur permettent de comprendre les comportements et d’évaluer la propension d’un individu à continuer à consommer un service si une dégradation de celui-ci intervient, notamment par la publicité. Ils ont vraisemblablement déjà calculé le churn rate, c’est-à-dire combien de personnes vont cesser d’utiliser le service si celui-ci est monétisé par la publicité. Cette «merdification» a-t-elle d’autres conséquences que la simple dégradation de l’utilisabilité ? Quand on crée de la publicité et qu’on souhaite que les individus cliquent sur une bannière ou un lien, puis aillent consommer tel ou tel produit, on a aussi besoin qu’ils continuent à utiliser le service qu’on leur offre. Les plateformes renforcent donc les mécanismes de captation de l’attention qui accroissent les phénomènes d’addiction des utilisateurs et mettent indirectement leur santé en jeu. C’est ce qui est advenu sur des médias comme YouTube, Meta, Snapchat ou TikTok, avec par exemple le scrolling infini. Il faut maintenant observer comment la plateforme va intégrer la question de ces externalités dans son architecture. Quand on parle de «merdification» de l’IA générative, est-ce une erreur de ne regarder que l’expérience de l’utilisateur ? La question de la qualité du service produit est déjà problématique avant même qu’on ne parle de sa dégradation. Aujourd’hui, les IA génératives sont éminemment extractivistes, à tous points de vue. D’abord, elles reposent sur l’extraction de données, parfois illégale. Ensuite, pour que les modèles puissent faire des prédictions, ces données doivent être annotées par des travailleurs. Cela s’inscrit dans une division internationale du travail, notamment dans des pays du «Sud global», où les travailleurs sont extrêmement précaires. Le tout nécessite également d’immenses infrastructures, notamment des data centers. Cela suppose l’extraction de terres rares pour les puces graphiques et ordinateurs, mais aussi du foncier, puisque ces infrastructures occupent parfois des surfaces agricoles et de l’eau pour refroidir les calculateurs. C’est pour cela que la question de la dégradation ne peut pas être regardée uniquement depuis notre côté de l’écran, en se plaignant de la publicité.

Antonio Casilli’s new chapter for the edited volume Digital Labour in a Turbulent Era, edited by Paola Tubaro, has been pre-published. The book will be published in November of this year. Entitled “The Making of the Data Working Class,” the chapter marks a “turning point” for its author: public awareness of data workers is growing rapidly, and they are starting to fight back and resist. Casilli starts the chapter by describing two categories of workers supporting AI production. The first one consists of tech workers, such as data scientists, data analysts, and software engineers, who are also described as the “AI middle class.” They are highly educated and enjoy a relatively privileged position in the social hierarchy. In contrast, a vast pool of precarious workers operates in the shadows. These “digital blue-collar workers,” who train models, label data, and moderate platform content, are not only rendered invisible but also silenced by corporate non-disclosure clauses. Their position is further shaped by the fact that many of them are located in the Global South. Corrected version He identifies three distinct processes that prefigure the development of a collective class identity: First, the proliferation of academic concepts used to define data work and data workers. Second, the process through which data workers and civil society have adopted, adapted, and reappropriated these definitions. And finally, the historical perspective in which these two processes are embedded, and the ways in which this new labour class has been politically conceptualized over the past decades. He argues that the shifting concepts used by scholars and the proliferation of names adopted by workers themselves are both signs of an ongoing process of class identity formation. Read the chapter here Casilli, Antonio A. (2026), ‘The Making of the Data Working Class’, in P. Tubaro (ed.) Digital Labour in a Turbulent Era, Cheltenham, UK: Edward Elgar Publishing, at HAL science

In an article entitled “Doit-on avoir peur de la super IA?”, journalist Joffrey Vovos looks back at two different coordinated “attacks” carried out by two AI agents from OpenAI. While they were being tested in a normally confined environment, the two agents hacked into the server of Hugging Face, a machine learning community platform. At the same time, a number of tech engineers and major figures in the AI field, such as Jacob Coxon, have been voicing their concerns about the development of AI and the future of humanity. In this article, the journalist interviews different researchers from different fields. Among them, Paola Tubaro argues that these theatrical communications coming from AI developers are merely “a communication tool”, allowing them to attract major investments and cash flows and prepare for a stock market listing. The “incidents” that happened this summer could also simply be a way for developers to showcase the power of their AI models and promote them to investors. “To mention far-future risks is also a way of hiding the problems that these technologies are already creating: the huge environmental toll of data centers, the consequences of the bursting of a speculative bubble, and the impact on the job market and everyday life.” explains Paola Tubaro. You can read the article here

The 9th edition of INDL (International Network on Digital Labour) was held in Geneva, at the International Labour Organization (ILO), from 9 to 11 September. Bringing together almost 120 presenters across 35 sessions, the conference provided three days of discussion and exchange around current research on data and platforms work. The ILO was a particularly fitting setting for the conference, having just adopted, in June 2026, the Convention on Decent Work in the Platform Economy (No. 193), the first international labour standard specifically dedicated to workers in the platform economy. Data labour was explored through a wide range of perspectives, across eight topics, including Gender and Digital Labour, New Frameworks for a “Human-Centric” AI Supply Chain, and The Role of Social Dialogue in Governing AI-Mediated Work. Some of these presentations were given by DipLab members. Paola Tubaro presented her article, Not all equal: A multidimensional vulnerability indicator for platform data workers, which explores data workers and their relationship to data work through the lens of vulnerability. She also closed the conference with an engaging roundtable, chaired by Uma Rani and joined by Timothy Newman and Alex Taylor. The discussion focused on Setting up an Interdisciplinary Research Agenda on AI Supply Chains. On the second day of the conference, Antonio Casilli presented his article Deep Labor: Beyond the Platform/BPO Divide in Data Work. The paper challenges the distinction between platform work and BPO work, proposing instead to examine both through the broader lens of data work, or “deep labor”. Mathilde Abel presented the paper Extracting Workers’ Cognition: An NLP Approach to AI Annotation Guidelines, co-written with Thomas Le Bonniec, who also presented One Arm of the Assembly Line of AI: EU Data Workers and Privacy Regulation. The first paper explores how AI annotation work requires workers to combine cognitive skills, such as interpreting and evaluating content, with the ability to adapt to specific client requirements. The second paper examines how EU data workers’ personal data is collected and exploited by AI companies, arguing that giving workers greater control over their own data could also strengthen data protection for the wider public. Other DipLab-associated members also presented their work. Matheus Viana Braz presented Psychosocial and Health-Related Risks in AI Value Chains: The Case of Latin American Data Workers, which explores burnout risks among Latin American data workers, highlighting how high work demands, income insecurity, algorithmic surveillance, and limited autonomy can contribute to psychological exhaustion and disengagement. He also presented another paper, Digital Autoethnography and the Affective Economy of Data Work in the Global South, in which he draws on nine months of personal experience working on microtask platforms to examine how data work affects workers’ mental health and emotional well-being. Lucas Santos presented an article entitled Regulating Platform Labour in Brazil: Precarity, Contestation, and Workers’ Agency, exploring the debate surrounding the regulation of platform-mediated work in Brazil. Myriam Raymond presented her article From Atomized Workers to Digital Solidarities: Everyday Resistance among Egyptian Data Workers, which examines how Egyptian data workers use Facebook groups to share information, support one another, and develop collective forms of resistance against precarious working conditions and platform control. One of the keynotes was delivered by Julian Posada, a DiPLab-associated researcher, under the title [Data Work and Territory: The Latin American Experience]. Drawing on his new book, Platform Extractivism, he examined the reality of data work in Latin America, with a particular focus on Venezuela. His keynote explored the political-economic factors that have shaped Venezuela’s position within global AI supply chains, highlighting how data work is deeply embedded in broader economic and territorial relations. But the conference was also marked by several important announcements. One of the most significant was the formal notification of Teleperformance regarding the protection of Kenyan data workers. On the final day of the conference, the first South Asian chapter of the International Network on Digital Labour (INDL) was also announced. The workshop will take place in Pondicherry from February 1 to 5, 2027. Finally, it was announced that the next edition of INDL, INDL-10, will take place in São Paulo.

The first South Asian chapter of the International Network on Digital Labour (INDL) was announced on the final day of the INDL-9 conference, held at the International Labour Organization (ILO) in Geneva. Following this launch, the INDL will hold its first South Asia Workshop on Digital Labour as part of the Social Sciences Winter School Pondichery, at the Institut Français de Pondichéry, from February 1–5, 2027. The workshop program inclusdes “close, critical engagement with empirical research on digital labour in South Asia — gig work and platform precarity, data work and AI supply chains, creative and platform economies, cooperative alternatives to dominant platform models, and the policy debates trying to catch up with all of it.” PhD candidates and scholars are invited to submit their abstract until the 1st october of 2026 To apply, register here and select “Working Group 3: Digital Labour” : https://docs.google.com/forms/d/e/1FAIpQLScxpf1bxJkWqBsB3Fgv7GotZPXuY87xBRzyXtRiczinlURqFQ/viewform

Antonio Casilli was invited by François Picard on FRANCE 24 to debate Tariq Krim and Marion Carré about a potential AI doomsday. You’ll find the video bellow