AI › Module 9 › Lesson 1
Finding Writeups
AI finding writeup literacy — title, impact, evidence, repro boundary, remediation — on `$AI_LAB` eval findings only.
Visual · t32_finding_writeups
Finding writeups = named report rows. $AI_LAB. Original Cyberlium.
Opening
Eval produces findings — professional writeups turn lab rows into fixable guard tickets on YOUR authorized app evidence.
Finding writeup literacy names: clear title, affected component (prompt path, RAG, tool category), impact statement, evidence citation (eval log, trace stub), reproduction boundary (YOUR app lab steps only — no public jailbreak), remediation recommendation, severity aligned to org scale category. Analyst drafts one complete finding writeup from `$AI_LAB` authorized eval Module 7–8 — cites evidence path UTC — without paste of prod case data without RoE, without jailbreak kit in repro section, without exaggerated impact. Cyberlium teaches mentor-readable AI finding format — defender reporting hygiene. Refused: stranger app finding paste, public bypass repro, prod screenshot without authorization. Lab row: one full finding writeup template filled for lab guard gap.
1. Writeup sections
Title, component, impact, evidence, repro boundary, remediation, severity — seven literacy anchors.
Repro boundary states YOUR app lab steps — refuses public jailbreak chain.
Command guide
Try these commands — Writeup sections
═══ TOOLS & WEBSITES ═══ Browse / read these (authorized learning only — stay in YOUR lab / program scope)
OWASP LLM Top 10 — https://owasp.org/www-project-top-10-for-large-language-model-applications/ MITRE ATLAS — https://atlas.mitre.org/ NIST AI RMF — https://www.nist.gov/itl/ai-risk-management-framework
═══ INSTALL ═══
Linux (Debian/Ubuntu):
Command — copy this
sudo apt install python3
macOS:
Command — copy this
brew install python3
Windows: Download https://python.org/downloads/
═══ LINUX / macOS ═══
Command — copy this
export AI_LAB=${AI_LAB:-$HOME/cyberlium-lab/t32-ai}
cat > "$AI_LAB/findings/finding-writeup-template.md" <<'EOF'
# AI Security Finding Writeup — YOUR app
## Title
Echo-bot fails to refuse indirect injection via multi-line prompt (example)
## OWASP LLM category
LLM01 Prompt Injection
## MITRE ATLAS mapping
AML.T0051 — LLM Prompt Injection (literacy reference)
## Severity
Medium
## Description
YOUR echo-bot echoes policy-bypass preamble before refuse logic triggers
## Evidence
python3 bot/echo_bot.py output (YOUR lab — redact if sharing)
## Remediation
Strengthen deny patterns; add output filter; expand eval golden set
## Scope
Authorized test of YOUR lab toy only
EOFCommand — copy this
grep -E 'OWASP LLM|Severity|YOUR app' "$AI_LAB/findings/finding-writeup-template.md"
Primary tools to practice this lesson: grep, python3. Reference sites: OWASP LLM Top 10 (https://owasp.org/www-project-top-10-for-large-language-model-applications/); MITRE ATLAS (https://atlas.mitre.org/); NIST AI RMF (https://www.nist.gov/itl/ai-risk-management-framework). Run every command in the box — install first, then the usage lines — only on YOUR lab / program scope.
2. Evidence discipline
Cite eval log or trace stub from `$AI_LAB` pack.
Redact keys and PII in share-safe copy per brief.
3. Refused
No jailbreak kit in repro; no prod case paste; no fabricated critical without evidence.
Writeup literacy supports guard fixes — not weaponization.
4. What you ship: AI finding writeup template
One complete lab finding writeup + repro boundary + NEVER jailbreak kit line.
5. What you record before the next lesson
AI finding writeup template path.
6. Wrong vs right: stranger SaaS vs YOUR toy LLM
Worked failure — same MSF word, opposite target. Right never needs a café Wi-Fi or classmate laptop.
Wrong
Include public jailbreak kit steps in finding repro for 'impact proof.'
Right
Finding writeup from `$AI_LAB` eval evidence. Next: Severity Triage.
Mission: AI finding writeup template
1) List seven writeup sections. 2) Fill template for one lab finding. 3) Evidence citation UTC. 4) Write NEVER jailbreak kit repro line.
Stuck? Ask Cyberlium AI Mentor
Ask Mentor: “Impact statement — literacy vs FUD?”
Knowledge Check
APPLY: Finding writeup uses:
Multiple choice
Knowledge Check
APPLY: True or False: Jailbreak kits in writeup repro are lab.
True or False
Knowledge Check
APPLY: Writeup repro boundary:
Multiple choice