The five risk categories — AI Midnight

Autonomy and loss of control, dangerous capabilities, real-world incidents, governance and safeguards, acceleration: definitions, scoring anchors and history.