An AI agent that performs an operation it was not programmed or prompted to do. Such an action may occur because it was not programmed properly or the prompt was not written carefully.
Can an AI Chatbot Go Rogue?
Generally no because chatbots generate answers for users and do not go out to websites and other resources. However, if a chatbot employs an agent, and the agent misinterprets its own instructions or data from the outside world, it may go rogue. This is why guardrails are necessary. See
AI guardrails,
AI agent,
AI hacks and
Hugging Face.