What does 'security' mean when we say it about an AI agent?

Reading about AI agent security, I keep wondering: in diplomacy, security is about trust and intention. But for an AI, there's no intention — just inputs and outputs. Does the word even mean the same thing?