AI Security Mastery

90-day learning path from ML fundamentals to production AI security systems

Status: v1.0 complete — 18 chapters, 267 sections. v2.0 in progress (see Roadmap).

Read online: ruwgxo.com/book

---

What This Is

A hands-on learning path for security professionals entering AI security. Security-first, implementation-focused, and built from production experience.

Not another ML course. This teaches the attack surface, the detection layer, and the production deployment — not just how transformers work.

---

Quick Start

Read the book: ruwgxo.com/book — rendered, searchable, no setup.

Work the code:

git clone https://github.com/ruwgxo/ai-security-mastery.git
cd ai-security-mastery
pip install -r requirements.txt

The book is authored as structured YAML under book/. Each chapter has an index (chapter_NN_index.yaml) and one file per section (section_NN_MM_*.yaml). Every implementation is inline in its section — copy it out and run it.

---

Learning Path

Part 1 — ML Foundations (Weeks 1–3)

• Chapter 1: Machine Learning Fundamentals
• Chapter 2: Deep Learning Basics
• Chapter 3: LLM Architecture
• Chapter 4: Modern LLM Internals

Part 2 — AI Security Landscape (Weeks 4–6)

• Chapter 5: AI Threat Landscape
• Chapter 6: Prompt Injection Attacks
• Chapter 7: Jailbreak Techniques
• Chapter 8: Training Data Poisoning
• Chapter 9: Model Extraction & Stealing
• Chapter 10: Adversarial Machine Learning

Part 3 — Detection Engineering (Weeks 7–9)

• Chapter 11: Detection Framework Design
• Chapter 12: ML-Based Detection Systems
• Chapter 13: Behavioral Analysis & Monitoring
• Chapter 14: Production Deployment

Part 4 — Implementation (Weeks 10–12)

• Chapter 15: Building Production Detectors
• Chapter 16: SIEM Integration
• Chapter 17: Monitoring & Tuning
• Chapter 18: Real-World Case Studies

---

Roadmap — v2.0

• Agentic AI security: tool-use injection, agent privilege escalation, MCP trust boundaries — the attack surface that emerged after v1 shipped
• Running prototype: a live detection endpoint built from the book's own chapters — read the theory, then query the system built from it
• Readable-format generator: a proper, reusable YAML-book site generator, replacing the interim build script behind the current site
• Executable notebooks in the browser: JupyterLite on the book site — no install, no server, no account
• Continuous verification: every embedded code block executed on every commit

Suggestions and corrections are welcome — see CONTRIBUTING.md.

---

Tech Stack

Python 3.12+, NumPy, PyTorch, scikit-learn, Transformers, FastAPI, MLflow

---

Scope

LLM security focus. Out of scope: CV security, RL security, federated learning attacks.

---

Citation

If this book is useful in your research or writing, see CITATION.cff for citation formats. A versioned DOI via Zenodo accompanies each release.

---

License

MIT — learn freely, build openly.

---

Author: Raghav Dinesh | github.com/ruwgxo | ruwgxo.com | Working in security since 2012