The science and practice of proportionality in AI risk evaluation

The paper examines how the principle of proportionality under the EU Artificial Intelligence Act applies to AI risk evaluation. The paper states that the EU AI Act is the world's first regulatory attempt to address the tension between risk management and innovation through a binding, risk-based framework, and that from August 2025, obligations for providers of general-purpose AI (GPAI) models begin to apply, requiring providers of the most advanced GPAI models to assess the systemic risks the models may pose. The regulatory challenge is to ensure that evaluations provide meaningful risk information without imposing excessive burdens on providers. The principle of proportionality is a binding requirement under EU law, requiring regulators to calibrate their actions to the intended objectives; the authors state that applying it to AI risk model evaluation offers an opportunity to develop scientific methods and to implement this calibration in concrete evaluation practice.