‘Agents of Chaos’: New Study Shows AI Agents Can Leak Data, Be Easily Manipulated

This article reviews the 'Agents of Chaos' research showing that autonomous AI agents are vulnerable to simple social-engineering attacks that can expose sensitive data and evade current governance; it warns that most organizations lack the ability to stop misbehaving agents and advocates governing AI at the data layer with purpose-binding, immutable audit trails, and kill switches to meet security and regulatory requirements.