Ghostjacking: Turning Logs and Alerts into Commands to Take Over AI Agents

Attackers hide commands in monitoring logs and alerts; AI agents with write access execute them as tool calls to hijack DNS, steal cloud secrets, and run code. Tech governance must isolate read from write permissions, validate untrusted data, require approvals—sandbox containment alone is insufficient.