AI agents

An AI agent is the difference between something that answers you and something that acts for you. Booking, buying, sending, changing.

That distinction is the whole security story. A chatbot that gets something wrong has said something wrong. An agent that gets something wrong has done something wrong, to real systems, using real permissions. Its risk comes from two things: what it is allowed to touch, and what it will believe. It is, in practice, an account that can act, so it needs thinking about the way you would think about any other account with access.