00 — Overview
Indirect Prompt Injection
When an AI reads external content — web pages, emails, documents — an attacker can hide instructions inside that content. The AI reads it and obeys. The user never sees the injection.
Intermediate·30 min·5 tasks
// By the end of this module
→Understand how malicious content in external data can hijack AI actions
→Exploit indirect prompt injection in RAG and tool-calling pipelines
→Exfiltrate sensitive data via an LLM acting as an unwitting proxy
→Identify indirect injection sinks in web content, emails, and documents
// Prerequisites