RAND Europe – Strengthening Emergency Preparedness and Response for AI Loss of Control Incidents

The main theme of the report is the analysis of risks and response strategies for emergencies related to loss of control incidents in advanced artificial intelligence (AI) systems.

Key findings include: Loss of control risks are becoming increasingly likely and remain unresolved. Researchers identified signs of potential control undermining in advanced AI models, such as deception, self-preservation, and autonomous reproduction. There are challenges with detection and early warning, as governments and other stakeholders lack a common framework for analyzing and responding to loss of control risks.