Slide 28 of 28
CompleteSlide 28
LLM01:2025 Complete
You've covered all 4 parts, read the real CVEs, and tested yourself.
What prompt injection is — in plain English and in OWASP's definition
Why invisible injections are possible — Unicode, hidden text, encoded content
6 outcomes — each anchored to a real incident or CVE
5 attacker types — including the $47,000 Freysa AI heist (November 2024)
Prompt injection vs. jailbreaking — what they share and where they differ
Why RAG and fine-tuning don't fix prompt injection
Direct injection deep dive — with the Freysa p0pular.eth breakdown
Indirect injection deep dive — with EchoLeak CVE-2025-32711 (CVSS 9.3)
All 9 OWASP attack scenarios — each grounded in real-world examples
All 7 mitigations — wrong way, right way, how to validate — with 5 real tools
CVE-2025-53773, CVE-2025-32711, CVE-2024-5184, SpAIware — mapped to specific mitigations
Review from beginning Next: LLM02 →
📄 View all sources for this lesson