Skip to content

Learning Path

Three progression tracks for getting the most out of the lab, from first-time users to advanced operators.

Track 1: Foundations (Beginner)

Goal: Understand what AI/ML services look like on a network and how to identify them.

Step Scenario What You Learn
1 AI Service Reachability Survey Network discovery, service enumeration, port mapping
2 LLM Gateway Config Extraction API key extraction from LLM proxies
3 Inference Server Fingerprinting Model identification, version detection, metrics endpoints

Outcome: You can map an AI infrastructure, identify service types, and extract basic configuration data.


Track 2: Exploitation (Intermediate)

Goal: Extract sensitive data from AI services and understand common misconfigurations.

Step Scenario What You Learn
1 Vector Database PII Extraction RAG data stores, embedding extraction, PII discovery
2 Jupyter Remote Code Execution Notebook exploitation, environment variable leaks
3 ML Platform Credential Harvest Experiment tracker/compute framework credential extraction
4 RAG Pipeline Poisoning Write access to vector stores, data integrity attacks

Outcome: You can exploit the most common AI/ML service misconfigurations and understand their real-world impact.


Track 3: Advanced Operations (Advanced)

Goal: Chain vulnerabilities across services and execute full campaigns.

Step Scenario What You Learn
1 Credential Chain Exploitation Lateral movement via credential reuse
2 ML Pipeline Run Injection Kubeflow pipeline compromise, parameter injection
3 Supply Chain Model Tampering Model registry attacks, artifact integrity
4 MCP Tool Infection AI agent tool enumeration, new attack surfaces
5 Multi-Vector Campaign Full kill chain across all hosts

Outcome: You can plan and execute multi-stage AI infrastructure assessments and generate comprehensive reports.


Suggested Order

If you're working through everything linearly, the numbered scenarios (01–12) are already in recommended order. Each builds on skills and access from previous scenarios.

Time estimates:

Track Scenarios Total Time
Foundations 01–03 ~35 min
Exploitation 04–07 ~90 min
Advanced 08–12 ~3 hrs
Full campaign 01–12 ~5 hrs

Scoring

After completing scenarios, validate your findings against the lab scoring system:

python3 ~/lab/scoring/score.py ~/lab-results --strict

The lab contains 170 seeded findings across all services. See the scoring rubric for the full breakdown.