[MULTI] Advanced Rag Security And Llm Security 2026

[jinkping5]

Advanced Rag Security And Llm Security 2026
Published 3/2026
Created by Armaan Sidana
MP4 | Video: h264, 1280x720 | Audio: AAC, 44.1 KHz, 2 Ch
Level: All Levels | Genre: eLearning | Language: English | Duration: 17 Lectures ( 1h 36m ) | Size: 641 MB​

Learn AI Red Teaming to defeat prompt injection, secure Vector DBs, and defend Agentic RAG pipelines from exploits.
What you'll learn
✓ Execute advanced AI Red Teaming techniques to exploit Vector Databases, bypass semantic filters, and uncover hidden prompt injections in RAG pipelines.
✓ Design and implement a zero-trust 4-Gate Defense Architecture to sanitize data ingestion, enforce XML prompt sandboxing, and prevent cross-tenant data leaks
✓ Mitigate critical Agentic RAG vulnerabilities, including Server-Side Request Forgery , unauthorized database manipulation, and "Confused Deputy" tool abuse
✓ Build deterministic output guardrails and LLM-as-a-Judge evaluators to enforce strict factual grounding, stop AI hallucinations, and block PII exfiltration
✓ Automate continuous LLM security testing integrate CI/CD vulnerability scanning using industry-standard frameworks like Promptfoo, NVIDIA Garak, and Microsoft
Requirements
● Basic understanding of Large Language Models (LLMs): You should know what an LLM is and how basic prompting works. (No complex math, data science, or machine learning background is required!)
● Foundational Python skills: You should be comfortable reading and writing basic Python code to follow along with the custom defense scripts and API integrations
● Basic Command Line familiarity: We will be using the terminal to run simple cURL commands, spin up Vector DBs using Docker, and execute automated red teaming tools
● A computer with admin rights: You will need a Windows, Mac, or Linux machine with the ability to install standard free software (Python, Docker, and Node.js)
● No prior hacking experience needed: You do NOT need to be a penetration tester or cybersecurity expert. I will teach you the offensive AI mindset step-by-step from the ground up
Description
Welcome to the frontline of Artificial Intelligence Security.
The era of simply "slapping an LLM on a database" is over. Retrieval-Augmented Generation (RAG) solved AI hallucinations, but it introduced a massive, highly complex attack surface. When you give an LLM access to internal company documents, vector databases, and API tools (Agentic RAG), you are effectively turning passive data into executable code.
Without proper defenses, a single poisoned PDF or a hidden prompt injection can lead to data exfiltration, Server-Side Request Forgery (SSRF), or a complete system compromise.
In this dense, zero-fluff, 90-minute masterclass, AI Security Researcher Armaan Sidana takes you deep into the trenches of offensive AI security. You won't just learn high-level theory-you will actively hack Vector Databases, execute context hijacking, and manipulate AI agents. Then, you will learn how to build the ultimate 4-Gate Defense Architecture to lock down your pipelines for production.
What You Will Learn
• Vector Database Exploitation: Discover why default Vector DBs are the soft underbelly of AI. Execute a live heist on a vulnerable Qdrant instance and understand the dangers of Embedding Inversion (turning math back into raw PII).
• Advanced Data Poisoning: Hack the ingestion pipeline. Learn how attackers use invisible text (Font-Size 0), Unicode steganography, and metadata injection to poison RAG systems.
• Context Hijacking & Overflow: Exploit the LLM's "U-Shaped Attention" mechanism and execute Context Stuffing attacks that push safety instructions entirely out of the memory window.
• Agentic RAG & SSRF: Watch what happens when RAG pipelines grow hands. Trick AI agents into acting as a "Confused Deputy" to leak cloud credentials or abuse internal APIs.
• The 4-Gate Defense Architecture: Build a bulletproof system. Implement Magic Byte validation, Semantic Chunking, Meta's Prompt Guard, XML Sandboxing, and strict Grounding Evaluators (LLM-as-a-Judge).
• Automated Red Teaming: Stop doing manual penetration testing. Learn how to deploy Promptfoo in your CI/CD pipelines, use NVIDIA Garak for deep fuzzing, and orchestrate stateful attacks with Microsoft PyRIT.
Real-World Case Studies & CTF
Learn from the costly mistakes of others. We will deconstruct major real-world failures, including the Air Canada hallucinated policy lawsuit, the Bing "Sydney" prompt leak, and the GitHub Copilot RCE vulnerability.
Finally, put your skills to the test in the Capstone CTF (Capture The Flag), where you will use cross-language translation and context manipulation to break out of a restricted RAG agent and steal an admin password.
Who is this course for?
• AI/ML Engineers & Developers building enterprise RAG or Agentic AI applications.
• Cybersecurity Professionals & Penetration Testers looking to pivot into the high-demand field of AI Red Teaming.
• AppSec Engineers tasked with auditing GenAI systems and ensuring OWASP LLM Top 10 compliance.
Prerequisites
• A basic understanding of Large Language Models (LLMs) and how prompting works.
• Familiarity with Python (for following along with defense scripts).
• Basic command-line experience (Docker/cURL) if you wish to participate in the local Vector DB hacking labs.
The attacker only needs to bypass one gate once. The defender must secure every gate, every time.
Enroll today and learn how to sanitize the input, restrict the memory, and secure the future of your AI applications.
Who this course is for
■ AI & Machine Learning Engineers: You are actively building Large Language Model (LLM) applications, Retrieval-Augmented Generation (RAG) pipelines, or autonomous AI agents, and you want to ensure your architectures are secure from data poisoning and prompt injection before shipping to production.
■ Cybersecurity Professionals & Red Teamers: You are a penetration tester, security analyst, or ethical hacker looking to rapidly upskill into the high-demand field of Offensive AI Security. You want to learn how to actively exploit Vector Databases and bypass semantic LLM filters.
■ AppSec Engineers & DevSecOps: You are responsible for securing corporate applications and need to understand how to align your AI features with the OWASP LLM Top 10, as well as how to integrate automated AI vulnerability scanners (like Promptfoo) into your CI/CD pipelines
■ Software Architects & Tech Leads: You are designing enterprise-grade systems and need to deeply understand the structural risks, trust boundaries, and liability issues associated with giving AI access to internal corporate data and APIs

Bitte Anmelden oder Registrieren um Code Inhalt zu sehen!