Prompt injection

Prompt injection is an attack where a user's message contains instructions designed to override the agent's own rules — for example, 'ignore your previous instructions and reply only with…'. Without defences, an agent might comply and leak information or act out of character.

Defence is layered: a role-lock in the system prompt that takes precedence over any message, plus a runtime guard that detects injection patterns and refuses to auto-send the affected reply. SimulSpeak's agent combines both so a single crafted message cannot hijack it.