Updating Purpose Limitation for AI – a normative approach to AI regulation

About

In this talk, we present our interdisciplinary approach to AI regulation by updating purpose limitation for AI. We address a critical blind spot in EU digital legislation: the secondary use of anonymised training data and pre-trained AI models. These forms of reuse–currently outside the scope of regulatory frameworks such as the GDPR and the AI Act–pose significant risks to individuals and society, including discrimination, erosion of public trust and the exacerbation of power asymmetries.

We argue that a shift is needed from regulating data processing alone to regulating the lifecycle of AI models. Our proposal introduces a dual framework of purpose limitation—applicable to both training datasets and trained models—to restrict the reuse of AI systems beyond their original ethical and political justifications. This approach aims to close a loophole that allows models to be transferred across contexts without democratic oversight or accountability.

Building on our previous work on purpose limitation, we will further explore how to practically implement this framework within the existing EU digital regulatory landscape. This includes governance mechanisms such as integrating purpose limitation with the AI Act’s model database and developing criteria for a tiered system to register, monitor, and, where necessary, restrict the purposes for which AI models are used.

Paper (OA): https://academic.oup.com/ijlit/article/doi/10.1093/ijlit/eaaf003/8121663

Project: https://purposelimitation.ai/

Rainer Mühlhoff is a philosopher and mathematician, leading the Ethics and Critical Theories of Artificial Intelligence research group at the University of Osnabrück, Germany. Rainer is also an associate researcher at the Einstein Center Digital Future, Berlin, and the Weizenbaum Institute, Berlin. Rainer’s work focuses on ethics, data protection and critical social theory in the digital society, critically exploring the societal impact of digital technologies through the lens of power relations and subjectivity. Recent publications: The Ethics of AI: Power, Critique, Responsibility (Bristol University Press, 2025); KI und der neue Faschismus (Reclam, 2025, forthcoming). Further information: https://RainerMuehlhoff.de/en/

Hannah Ruschemeier is Professor of Public Law, Data Protection Law, and the Law of Digitalisation at the University of Hagen. She is an associated researcher at the Center for Advanced Internet Studies (CAIS NRW), at the research centre for digitale_kultur, a board member of RAILS e.V., and an editor of the journal LTZ – Legal Tech. Her research combines traditional questions of public law with the challenges of legal transformation and an interdisciplinary perspective.