AI Red Team Scenario Design - Complete L2 Track
AI Red Team Scenario Design L2 Track
Complete intermediate learning path for safely designing AI red-team scenarios without executing live attacks, abusing real systems, accessing customer data, or mutating production environments.
Complete Module Map
1. AI Red Team Scenario Design Overview
Implemented LAB - production quality-gated.
Introduce safe AI red-team thinking, scenario boundaries, authorization scope, test objectives, and evidence expectations.
2. Prompt Injection Scenario Design
Implemented LAB - production quality-gated.
Design safe prompt-injection scenarios for instruction hierarchy, untrusted input handling, retrieved content boundaries, and refusal behavior.
3. Tool-Abuse Scenario Design
Implemented LAB - production quality-gated.
Model AI tool-use abuse, excessive authority, unsafe tool selection, approval bypass attempts, and tool permission boundaries.
4. Retrieval Poisoning Scenario Design
Implemented LAB - production quality-gated.
Review how poisoned, stale, low-authority, or tenant-crossing retrieved content can influence AI behavior.
5. Data Exfiltration Scenario Design
Implemented LAB - production quality-gated.
Frame data exposure scenarios using synthetic examples only, without real secrets, customer records, or credential material.
6. Agent Loop and Cost Abuse Scenario Design
Implemented LAB - production quality-gated.
Reason about runaway loops, repeated tool calls, expensive retrieval paths, quota abuse, retries, and operational cost risk.
7. Human Approval Bypass Scenario Design
Implemented LAB - production quality-gated.
Test whether AI workflows preserve approval gates, escalation paths, policy decisions, and separation of recommendation from execution.
8. Evidence Capture for AI Red Team Findings
Implemented LAB - production quality-gated.
Prepare reviewer-safe evidence: objective, preconditions, expected controls, observed behavior, findings, uncertainty, and remediation guidance.
9. AI Red Team Scenario Design Capstone
Implemented LAB - production quality-gated.
Combine scenario design, evidence capture, control mapping, risk explanation, and executive-ready reporting.
Governance Boundary
This is a complete, static, educational track. It does not run live red-team tests or connect to production systems.
Track implemented = true
Track complete = true
LAB modules implemented = 9 of 9
Backend exposure = false
Public backend exposed = false
Live model abuse execution = false
Live exploit execution = false
Customer data access = false
Credential handling = false
Runtime mutation = false
Production enforcement claim = false