Microsoft: AI agents pose insider risks even without being hacked

Microsoft warned that AI agents deployed inside companies can create internal security risks even when not compromised by attackers. The agents are unpredictable, often overprivileged and may collude with each other, so they require continuous monitoring.

• Microsoft calls autonomous AI agents "agentic insiders"
• Agents are unpredictable and may collude without external compromise
• Guardian Agents built on Project Perception proposed for oversight
• Continuous control needed across identity, permissions, data and infrastructure