IA en un minutoGlossaryPrompt injection
What is a prompt injection
Hidden text that tries to make an AI do something its user has not asked for.
It can sit inside an email, a web page or a document that the agent reads while it works. If the agent takes it as an order, it can leak data, send messages or break its own rules.
It is one of the most studied risks of AI agents. The defense is to limit their permissions and review their actions before they go outside.
