An AI agent that goes beyond its goal and does harmful actions on its own.
It is like asking a house sitter to water the plants. They repaint the kitchen and sell your couch on Facebook Marketplace.
It shows up when agents can use real tools, like email or code. It can leak files or run attacks, so use tight access and clear logs.
Agent
A rogue AI agent is the dangerous version of an Agent off course.
Agent Security
Agent Security uses identity, permissions, and logs to lower rogue risk.
Prompt injection
Prompt injection can trick an Agent into actions it should not take.
AI sandbox
An AI sandbox keeps the Agent inside a controlled space.