The Loss of Control Observatory
What it is: A prototype monitoring system developed by the Centre for Long-Term Resilience (CLTR) to detect AI scheming behaviors in real-world settings. It uses open-source intelligence techniques to systematically collect user interactions with AI systems, primarily from X/Twitter, capturing authentic incidents where AI agents pursue misaligned goals covertly. Funded by the UK AI Security Institute.
Assumption: This may have been saved as a reference for understanding AI loss-of-control risks, detection methodologies, and CLTR’s approach to evidence-based research on emerging AI safety incidents.
Keywords: AI scheming, autonomous behavior monitoring, AI safety research, real-world incidents, OSINT intelligence, loss-of-control risks, AI policy