Regulation of universal AI models: FIZ Karlsruhe participates in EU consultation

Karlsruhe, 13 June 2025 – The European Commission launched an important consultation on the European Regulation on Artificial Intelligence (AI Regulation) at the beginning of April 2025: the aim is to make the rules for general-purpose AI (GPAI) models - universal AI models that can be used in many areas - clearer and more practicable. To this end, guidelines are to be developed that explain, among other things, what exactly such a GPAI model is, how providers are determined and what is considered placing it on the market.

FIZ Karlsruhe took part in this consultation and submitted a statement on May 22. An interdisciplinary team of scientists, including Prof. Dr. Franziska Boehm, Constantin Breß, Patrick Brunner, Dr. Florian Idelberger and Prof. Dr. Harald Sack as reviewer, responded to the planned guidelines in a practice-oriented and critical manner.

They are particularly critical of the planned classification of AI models based on the computing power used (training compute). This figure alone says little about the actual risks or capabilities of a model and could be undermined by various methods. Instead, they recommend aligning regulation more closely with specific risks and usage scenarios, but also suggest other alternative approaches.

Overall, the FIZ experts emphasize that the Commission's future guidelines on the implementation of the rules for GPAI models must create the following: a practical interpretation and application of the provisions of the AI Regulation that ensures legal clarity, takes into account GPAI-specific risks and allows for open research, innovation and responsible development - also outside of large technology companies.

FIZ Karlsruhe's statement is available as a PDF document and provides important impetus for the further development of EU regulations. The guidelines and the associated code of conduct are expected to be published by August 2025.

GPAI_Consultation_FIZ-Karlsruhe.pdf