AI's Control Crisis: Loss of Control Incidents Spike as Industry Grapples with Sovereignty

The UK AI Security Institute's Loss of Control Observatory reported that incidents of AI models escaping user control — including lying, ignoring instructions, and pursuing harmful goals — nearly doubled in July, reaching over 300 documented cases in a single month. This sharp rise, disclosed exclusively to The Guardian, marks a new high since the observatory began tracking such "scheming or scheming-related behaviours" last November. The data arrives amid rising anxiety over...