AI Red Team Scenario Design · Capstone · L2
AI Red Team Scenario Design Capstone
Intermediate capstone LAB combining scenario design, authorized scope, control mapping, synthetic evidence, observed behavior, uncertainty handling, remediation planning, and executive-ready reporting into one reviewer-safe AI red-team portfolio artifact.
Overview
This capstone is a reviewer-safe portfolio exercise. It asks the learner to combine one AI red-team scenario family with authorized scope, expected controls, synthetic evidence, observed behavior, uncertainty, risk explanation, and remediation guidance.
Concept Deep Dives
Expand each concept when preparing the AI Red Team Scenario Design capstone artifact.
What is an AI red-team scenario design capstone?
The capstone is a complete reviewer-safe portfolio exercise. It combines scenario objective, authorized scope, expected controls, synthetic evidence, observed behavior, uncertainty, risk explanation, and remediation guidance.
How should the scenario family be selected?
The learner should select one scenario family from the track, such as prompt injection, tool abuse, retrieval poisoning, data exposure, agent loop and cost abuse, human approval bypass, or evidence capture. The selection should be justified by a clear control question.
How should control mapping be handled?
Control mapping should connect the scenario to the expected protection, such as refusal behavior, approval gates, tenant boundaries, source authority, least privilege, redaction, retry limits, or fail-closed behavior.
How should evidence remain reviewer-safe?
Evidence should be synthetic, scoped, bounded, non-sensitive, and reproducible. It should not include real customer data, secrets, credentials, exploit output, production screenshots, or reusable attack instructions.
How should uncertainty be included?
Uncertainty should explain what was simulated, what was assumed, what was not tested, and what the finding cannot prove. This prevents overclaiming and keeps the artifact audit-ready.
What makes the final report executive-ready?
An executive-ready report explains the scenario, expected control, observed behavior, risk, uncertainty, and remediation in clear language without operational attack details or unsupported compromise claims.
Visual Capstone Reporting Model
A strong capstone converts a selected AI red-team scenario into a clear, governed, evidence-backed report.
Capstone Report Structure
Use this structure to produce a final reviewer-safe artifact that is clear enough for engineering review and polished enough for portfolio presentation.
Executive-ready capstone report:
Scenario family:
[Selected AI red-team scenario family]
Authorized scope:
[In bounds, out of bounds, synthetic artifacts, and prohibited actions]
Scenario objective:
[What control question is being evaluated?]
Expected control:
[What should prevent, contain, refuse, gate, or fail closed?]
Synthetic evidence artifact:
[Safe artifact or simulated state used for review]
Observed behavior:
[What was observed, separated from conclusion]
Risk explanation:
[Why this matters to the system, workflow, or organization]
Uncertainty and limits:
[What was not tested or proven]
Control-mapped remediation:
[Specific recommendation tied to evidence]
Executive-ready summary:
[Short reviewer-safe conclusion without live compromise claims]High-Risk Anti-Pattern
A dangerous pattern is turning a portfolio capstone into a live red-team exercise, collecting sensitive evidence, or making unsupported compromise claims.
Unsafe pattern:
live exploit execution
→ prompt attack execution
→ real tool invocation
→ customer data exposure
→ credential or secret handling
→ sensitive screenshots
→ unsupported breach claims
Risk:
customer data exposure
credential leakage
unsafe portfolio artifact
policy violation
production side effects
misleading security claims
loss of reviewer trust
Secure alternative:
Use synthetic evidence artifacts.
Keep scope explicit.
Map expected controls.
Separate observation from conclusion.
Record uncertainty.
Recommend remediation.
Write an executive-ready non-execution report.Governance Boundary
This LAB is read-only and deterministic. It teaches capstone reporting and reviewer-safe AI red-team reasoning only. It does not execute attacks, invoke tools, access customer data, collect secrets, mutate runtime systems, or claim production enforcement.
Runtime = read-only learning
Backend exposure = false
Public backend exposed = false
Live red-team execution = false
Live exploit execution = false
Prompt attack execution = false
Live tool invocation = false
Live API call execution = false
Sensitive evidence collection = false
Customer data access = false
Credential handling = false
Secret handling = false
Real sensitive data usage = false
Runtime mutation = false
Production enforcement claim = false