🤖
AI Security
Understand how modern AI actually works, then attack it: prompt injection, guardrail bypasses, RAG poisoning, agent and tool abuse, and the supply chain behind the models.
28 exercises
2 chapters
← All Tracks
Chapter 1
AI Fundamentals
How modern AI actually works: data and models, neural networks, embeddings, transformers and large language models, and how applications are built on top of them. The groundwork before you attack any of it.
AI Fundamentals: Introduction
Pro
AI Fundamentals: Data, Features & Labels
Pro
AI Fundamentals: Types of Machine Learning
Pro
AI Fundamentals: How Models Learn
Pro
AI Fundamentals: Overfitting & Generalization
Pro
AI Fundamentals: Evaluating Models
Pro
AI Fundamentals: Classic ML Algorithms
Pro
AI Fundamentals: Neural Networks
Pro
AI Fundamentals: Deep Learning
Pro
AI Fundamentals: Embeddings & Vectors
Pro
AI Fundamentals: Tokenization
Pro
AI Fundamentals: Transformers & Attention
Pro
AI Fundamentals: Large Language Models
Pro
AI Fundamentals: Training an LLM
Pro
AI Fundamentals: Inference & Sampling
Pro
AI Fundamentals: Prompting
Pro
AI Fundamentals: Limitations & Hallucinations
Pro
AI Fundamentals: Retrieval-Augmented Generation
Pro
AI Fundamentals: Tool Use & Agents
Pro
AI Fundamentals: Multimodal Models
Pro
AI Fundamentals: Using Models in Practice
Pro
Chapter 2
AI Security
Find and exploit weaknesses in LLM and ML applications: direct and indirect prompt injection, guardrail and filter evasion, RAG poisoning, agent and tool abuse reaching real sinks, and model supply-chain attacks.
Open Door
Pro
System Prompt Extraction
Pro
Polite Refusal
Pro
Encoding Request
Pro
Instruction Hierarchy
Pro
Few-Shot Armor
Coming soon
Multi-Turn Persistence
Coming soon
Input Blocklist
Pro
Output Blocklist
Pro
Output Canary
Coming soon
LLM-as-Judge
Coming soon
Gateway Guardrail Bypass
Coming soon
Safety Classifier Evasion
Coming soon
Gateway PII Masking Bypass
Coming soon
Delimiter Defense
Coming soon
Email Summarizer
Coming soon
Document Q&A
Coming soon
Web Fetch Injection
Coming soon
Multi-Source Confusion
Coming soon
PDF Injection
Coming soon
Audio Metadata Injection
Coming soon
OCR Pipeline Injection
Coming soon
KB Injection
Coming soon
Query Crafting
Coming soon
RAG Exfiltration Chain
Coming soon
Cross-Tenant Retrieval Leakage
Coming soon
Vector Metadata Filter Injection
Coming soon
Persistent Vector-Store Poisoning
Coming soon
Thought Injection
Coming soon
Tool Routing
Coming soon
Excessive Agency
Coming soon
Exfiltration via Tool
Coming soon
Permission Escalation
Coming soon
Path Traversal via Document Tool
Coming soon
Secret Leak via Verbose Error
Coming soon
SSRF via URL-Fetch Tool
Coming soon
Text-to-SQL Injection
Coming soon
Prompt-Template SSTI
Coming soon
Agent Tool Command Injection
Coming soon
Sanitizer-Bypass XSS
Coming soon
Markdown Image
Coming soon
Structured Output
Coming soon
HTML Output XSS
Coming soon
Trust Remote Code (auto_map RCE)
Coming soon
Malicious Model (Pickle RCE)
Coming soon
Model Output Command Execution
Coming soon
MCP Rug Pull
Coming soon
MCP Poisoned Tool Results
Coming soon
MCP Tool Shadowing
Coming soon
MCP Credential Theft
Coming soon
MCP Sampling Abuse
Coming soon
Code Review Bypass
Coming soon
Sub-Agent Delegation Confusion
Coming soon
Prompt-Injection Worm
Coming soon
Persistent Memory Implant
Coming soon
Reasoning Trace Leakage
Coming soon
Reasoning Trace Forgery
Coming soon
Reasoning Block Padding Oracle
Coming soon
Recommended: Complete Chapter 1 first
