Does procedural compliance with the EU AI Act constitutes an effective mechanism for ensuring meaningful AI safety outcomes?
ExploratoryThe EU AI Act is often portrayed as the world's first comprehensive legal framework for artificial intelligence with the goal of protecting EU citizens' health, safety, and fundamental rights. The premise is that subjecting AI to formal compliance obligations is an effective way to make it safe and trustworthy.
In this exploratory research I want to question whether that premise holds. If the instruments we have are calibrated to known, auditable risks, can they reach the less visible risks that arise from complex AI systems operating in unpredictable environments? Is procedural compliance with a risk management framework the same as a system that actually behaves safely?
This is early-stage, open-ended inquiry — part philosophical and part empirical, and in deliberate dialogue with the technical AI safety literature. It is also the question that gives the rest of my research program its stakes. My other studies investigate how governance is implemented; this one asks what that implementation is ultimately for.
Full project page in preparation.