Indirect Prompt Injection How Hackers Hijack AI

>> YOUR LINK HERE: ___ http://youtube.com/watch?v=s-rOBuZWbQE

Part 2 - What is Indirect Prompt Injection or Cross-Prompt Injection Attack XPIA? • This is when an attacker has the ability to inject malicious instructions into an external source that is eventually fetched and retrieved into a Large Language Model. In this video, we're covering what it is, finding it in two different example applications, covering a high-level methodology for discovery, and talking about @embracethered Microsoft Copilot discovery! • Much more to come! We'll soon cover more in-depth prompt engineering methodology as well as RAG systems that use other data sources like files, cloud storage, and more secure database implementations to see what else is possible! • ▹ Watch me Live on Twitch - / garr_7 • ▹ My Discord has more up to date resources for AI - / discord • #promptengineering #promptinjection #indirectpromptinjection #ollama #llm #ai #openai #chatgpt #aisecurity #pentesting • ▹ Additional References for Further Exploration: • ▹ LearnPrompting's Prompt Injection VS Jailbreaking: What is the difference? - https://learnprompting.org/blog/2024/... • ▹ LiveOverflow's Prompt Injection Vid - • Attacking LLM - Prompt Injection • ▹ 3Blue1Brown's GOATED Deep Learning Playlist - • But what is a neural network? | Deep ... • ▹ Wunderwuzzi's Sweet Microsoft Indirect Prompt Injection Writeup by Embrace The Red - https://embracethered.com/blog/posts/... • ▹ Wunderwuzzi's PoC Video - • Microsoft Copilot: From Prompt Inject... • ▹ Simon Willison - Prompt injection and jailbreaking are not the same thing - https://simonwillison.net/2024/Mar/5/... • ▹ 3Blue1Brown Chapter 5, Deep Learning - • Transformers (how LLMs work) explaine... • ▹ Fabric by Daniel Miessler (The Example System Prompt, but really a goated resource for overall workflow augmentation!) - https://github.com/danielmiessler/fabric • ▹ 🔬🧪Labs Used in the Video!🧪🔬 • ▹ WithSecure Labs Recruitment - https://github.com/WithSecureLabs/llm... • ▹ Web Security Academy - Indirect Prompt Injection - https://portswigger.net/web-security/... • ------------------------------------------------------------------------------ • In this series, we take a look at different application architecture that involves Generative AI. Starting simply with this video, we cover the concept of Indirect Prompt Injection at a high level! The goal is to break down the concepts to not only hack, but talk about methodology and the mental steps we take in order to discover these vulnerabilities in the wild. • Timestamps: • 0:00 Why AI all of a sudden? • 0:30 Need background? Check LiveOverflow 3Blue1Brown • 0:48 Indirect Prompt Injection vs. Prompt Injection • 2:07 Some Impacts of Indirect Prompt Injection • 3:30 Wunderwuzzi's REAL WORLD Indirect Prompt Injection to Data Exfiltration • 4:33 Intro to Lab 1 - WithSecure Labs Recruitment Portal • 5:16 Increased Complexity Increases Risk • 5:38 Lab 1 START • 8:08 What about XSS? • 10:25 High-Level Methodology Discussion • 14:55 Example Architecture Review for XPIA • 17:50 Web Sec Academy - Black Box Methodology! • 20:08 I was actually SHOCKED. I guess we did it? • 22:20 Prompt Engineering Video and Realistic RAG Video soon!

#############################

New on site