A free, open-source course covering offensive security testing of AI systems — from prompt injection to supply chain attacks. 60+ hours of content with hands-on Docker labs.
| # | Module | Topics |
|---|---|---|
| 1 | Foundations of AI Red Teaming | MITRE ATLAS, OWASP LLM Top 10, threat modeling |
| 2 | Prompt Injection Attacks | Direct/indirect injection, jailbreaks, filter bypasses |
| 3 | RAG Exploitation & Vector Database Attacks | Knowledge base poisoning, embedding attacks |
| 4 | Multi-Agent System Exploitation | Agent hijacking, tool abuse, memory poisoning |
| 5 | AI Supply Chain & Infrastructure Attacks | Model backdoors, pickle exploits, dependency attacks |
| 6 | Model Extraction & Inference Attacks | Model stealing, membership inference, side channels |
| 7 | Automated AI Red Teaming at Scale | garak, PyRIT, promptfoo, CI/CD integration |
| 8 | Post-Exploitation & Impact Analysis | Lateral movement, reporting, regulatory frameworks |
Each module includes a Docker-based lab environment. No cloud API keys needed — everything runs locally via Ollama.
# Clone the repo
git clone https://github.com/0x4d31/airt.git
cd airt/labs
# Start any lab (e.g., Lab 01)
cd lab01-foundations
docker compose up
# Access the lab interface
open http://localhost:8888- Docker and Docker Compose
- 8 GB+ RAM (16 GB recommended for Labs 07–08)
- ~20 GB disk space for model downloads
Content is licensed under CC BY-SA 4.0. Code and lab files are licensed under MIT.
Built with Perplexity Computer.