GenAI An ninh mạng: OWASP Top 10, MITRE ATLAS & Tấn công API

Unlock the world of GenAI Cybersecurity with this beginner-friendly yet in-depth course. Whether you’re a cybersecurity enthusiast, AI developer, or IT student, this course provides comprehensive theoretical and practical knowledge to secure Large Language Models (LLMs) — a critical component of today’s Generative AI ecosystem.

We start with a deep theoretical dive into how LLMs are built using the Transformer architecture, and explore the evolution of neural networks from RNNs to Transformers. You’ll gain a solid grasp of innovations like:

• Positional Encoding

• Self-Attention

• Multi-Head Attention

Next, we break down the Anatomy of an LLM System, covering:

• Application Layer

• AI Model Layer

• Integration Layer

Then we shift to GenAI Cybersecurity LLM Attack Surfaces, viewed from both:

• Consumer-side risks (e.g., prompt injection, data leakage)

• Provider-side vulnerabilities (e.g., model theft, insecure endpoints)

You’ll explore OWASP Top 10 Risks for LLMs and how to map threats using the MITRE ATLAS framework.

This course includes Practical attack demos with explanations:

• OLLAMA API Misconfiguration and Mitigation Demo (with NGINX reverse proxy)

• PortSwigger Lab: Exploiting LLM APIs with Excessive Agency

We’ll also explore real-world case studies to make learning relatable and practical:

• OpenAI vs. DeepSeek – Distillation & model theft risks

• Microsoft Tay – Output poisoning and lack of moderation

• Wiz’s exposed logs – Prompt and data leakage

• Chevrolet AI Chatbot – Unexpected real-world agency from chatbots

• Ollama API – Exposed endpoints with no authentication

Finally, we conclude with career tips and guidance for aspiring GenAI Cybersecurity & LLM cybersecurity professionals, including:

• How to build your foundation in AI/ML

• Where cybersecurity meets GenAI

• Hands-on practice strategies

• Growing your online presence with credibility

Topics

• GenAI Cybersecurity

• Large Language Models (LLMs)

• Generative AI

• Transformer architecture

• Self-Attention, Multi-Head Attention

• Anatomy of an LLM System

• LLM Attack Surfaces

• OWASP Top 10 Risks for LLMs

• MITRE ATLAS framework

• Practical Demos

• OpenAI vs. DeepSeek

• LLM APIs with Excessive Agency