What is indirect prompt injection?
It is an attack embedded in content the model retrieves or observes, such as a webpage, document, email, or tool result.
It is an attack embedded in content the model retrieves or observes, such as a webpage, document, email, or tool result.