Towards agentic security in the emerging European Union AI policy

The article analyses how the European Union (EU) constructs AI-related security in its emerging AI policy. It conceptually engages with the elements of riskification: a referent object and potential harm. By conducting a discourse analysis of selected documents and semi-structured expert interviews, both fundamental rights and a democratic political system emerge as referent objects, while intrusion, discrimination and AI autonomy are posited as conditions of potential harm to them. The analysis contributes to the debates on AI-related security by introducing the concept of agentic security. It focuses on protecting human agency, understood as the capacity to sustain control and decision-making power vis-à-vis AI. This security logic is driven by the imaginary of human-machine interaction, aiming to embed AI in the EU’s priorities, and refuses other constellations between human agency and machines. This position, grounded in anthropocentric views, seeks to establish AI governance that supports and protects a liberal order.