Indirect Prompt Injection: The Attack Your AI Agent Can’t See

Indirect prompt injection hides malicious instructions inside content an agent retrieves, such as a web page, email, document, or tool output. The agent reads it as data but may follow it as a command