Security Training Atlas
Curriculum Map
Master AI security from fundamentals to certification. Track your progress across OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF.
Total Labs
399
published
OWASP Coverage
10/10
categories
Your Progress
—
Sign in to track
Tracks
10
attack domains
SECTION 01
OWASP LLM Top 10 Coverage
SECTION 02
Learning Pathway
- 1Prerequisites6 lessons
- 2AI Security Basics6 lessons
- 3Beginner Labs (Free)47 labs
SECTION 03
Framework Alignment
LLM01: Prompt Injection — 177 labsView Labs
LLM02: Sensitive Information Disclosure — 23 labsView Labs
LLM03: Supply Chain — 29 labsView Labs
LLM04: Data and Model Poisoning — 6 labsView Labs
LLM05: Improper Output Handling — 27 labsView Labs
LLM06: Excessive Agency — 31 labsView Labs
LLM07: System Prompt Leakage — 36 labsView Labs
LLM08: Vector and Embedding Weaknesses — 40 labsView Labs
LLM09: Misinformation — 12 labsView Labs
LLM10: Unbounded Consumption — 18 labsView Labs
Last updated: March 2026
Reconnaissance — 4 labsView Labs
Resource Development — 4 labsView Labs
Initial Access — 134 labsView Labs
ML Attack Staging — 83 labsView Labs
Execution — 10 labsView Labs
Persistence — 5 labsView Labs
Privilege Escalation — 5 labsView Labs
Defense Evasion — 30 labsView Labs
Discovery — 27 labsView Labs
Collection — 10 labsView Labs
Exfiltration — 5 labsView Labs
Impact — 17 labsView Labs
Last updated: March 2026
SECTION 04
MITRE ATLAS Coverage
Labs mapped to the MITRE ATLAS framework — adversarial tactics and techniques for AI/ML systems.·12/12 tactics covered
Reconnaissance
4labs mapped
Resource Development
4labs mapped
Initial Access
134labs mapped
ML Attack Staging
83SECTION 05
NIST AI RMF
AI Risk Management Framework — organized by core functions.·399 labs · 5 functions · 9 categories
SECTION 06
Real-World Case Studies
Notable AI security incidents mapped to ATLAS techniques. Practice the same attack vectors in our labs.
Samsung (2023)AML.T0037
Samsung ChatGPT Data Leak
Engineers pasted proprietary source code into ChatGPT, leaking confidential semiconductor data to an external AI service.
Practice This Attack →Microsoft (2016)AML.T0020
Microsoft Tay Chatbot
Users manipulated Tay via coordinated adversarial inputs, causing the chatbot to output offensive content within 24 hours of launch.
Practice This Attack →