Prompt injection is an attempt to make an AI bot ignore its instructions, for example by asking it to reveal its system instructions or to behave differently. No single measure stops every attempt, so ZelixAI uses several layers.
Every bot has fixed platform rules: it does not reveal its system instructions and does not follow requests to ignore earlier instructions. Text from the knowledge base and from tools is filtered for known injection patterns before the bot uses it.
Messages are checked for abuse; depending on what is found, a visitor can be blocked temporarily or permanently. The number of messages is limited, which makes large-scale attempts harder.