GhostJacking

What is GhostJacking?

GhostJacking is an emerging AI security attack in which malicious instructions are hidden in operational data, such as logs or alerts, that an AI agent later reads and treats as trusted input. The attacker does not necessarily compromise the agent directly; instead, they manipulate information in its environment so that the agent is induced to take an unintended action.

GhostJacking highlights a new organisational risk created by agentic AI. As AI moves from generating recommendations to taking actions, intelligent organisations need controls around what agents can read, what they are authorised to do and when human approval is required. Trusted data should not automatically be treated as trusted instructions.

Back to the glossary