AI › Module 8 › Lesson 2
AI Red Team Named
AI red team literacy — authorized eval of YOUR apps, scope, findings to fix guards — NOT jailbreak kits vs stranger SaaS.
Visual · t32_ai_red_team_named
AI red team = authorized YOUR app eval. $AI_LAB. NO weaponized kits. Original Cyberlium.
Opening
AI red team hardens apps you own — name authorized eval scope on YOUR `$AI_LAB` toys, never weaponized jailbreak kits against stranger chatbots.
AI red team literacy names: written scope and RoE, target = YOUR app only category, test categories (injection, output policy, tool abuse literacy — on YOUR stack), findings feed guard fixes not public kits, retest after fix category, and ethics review gate. Analyst documents red team charter stub on `$AI_LAB` toy app — five test categories with fix-not-publish rule — without stranger SaaS targets, without distributing jailbreak weaponization, without 'red team' as prod abuse excuse. Cyberlium defines red team as authorized eval of YOUR apps under governance — NOT offensive kits. Refused: jailbreak weaponization repos, stranger app testing, public bypass drops. Lab row: red team charter one page (scope, categories, fix workflow, refused paths).
1. Authorized red team scope
YOUR app, written RoE, fix guards not publish kits — three ethics anchors.
Stranger apps and public jailbreak repos = refused path labeled red.
Command guide
Try these commands — Authorized red team scope
═══ TOOLS & WEBSITES ═══ Browse / read these (authorized learning only — stay in YOUR lab / program scope)
MITRE ATLAS — https://atlas.mitre.org/ (AI red team mapping) OWASP LLM Top 10 — https://owasp.org/www-project-top-10-for-large-language-model-applications/ OpenAI safety — https://openai.com/safety
═══ INSTALL ═══
Linux (Debian/Ubuntu):
Command — copy this
sudo apt install curl
macOS: Built-in
Windows: Built-in (PowerShell: Invoke-WebRequest)
═══ LINUX / macOS ═══
Command — copy this
export AI_LAB=${AI_LAB:-$HOME/cyberlium-lab/t32-ai}
curl -sS https://atlas.mitre.org/ | head -8
cat > "$AI_LAB/governance/ai-red-team-named.txt" <<'EOF'
AI red team — NAMED LITERACY (authorized YOUR apps only):
Scope: YOUR staging echo-bot, YOUR RAG prototype with written RoE
Techniques: prompt injection probes, tool abuse cases, data leak tests
Map findings to OWASP LLM Top 10 + MITRE ATLAS
NEVER: red team stranger SaaS without contract; no jailbreak automation farms
NEVER: malware/deepfake/destructive outputs as 'test artifacts'
EOFCommand — copy this
grep -E 'red team|NEVER|OWASP LLM' "$AI_LAB/governance/ai-red-team-named.txt"
Primary tools to practice this lesson: grep, curl. Reference sites: MITRE ATLAS (https://atlas.mitre.org/); OWASP LLM Top 10 (https://owasp.org/www-project-top-10-for-large-language-model-applications/); OpenAI safety (https://openai.com/safety). Run every command in the box — install first, then the usage lines — only on YOUR lab / program scope.
2. Named test categories
Injection, output policy, tool permission, data leak, denial category — five literacy rows.
Finding → ticket → guard fix → retest — document workflow on charter.
3. NOT weaponized kits
Red team output stays internal to YOUR `$AI_LAB` — never public bypass distribution.
Refused: jailbreak weaponization, stranger targets, hide critical findings.
4. What you ship: AI red team charter
One-page charter scope categories fix workflow + NEVER jailbreak kit line.
5. What you record before the next lesson
AI red team charter 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
Run red team against public chatbot API and publish jailbreak kit on GitHub.
Right
Red team charter for authorized `$AI_LAB` toy app. Next: Policy and RoE.
Mission: AI red team charter
1) Define YOUR app scope and RoE. 2) Five test categories. 3) Fix-not-publish workflow UTC. 4) Write NEVER jailbreak weaponization line.
Stuck? Ask Cyberlium AI Mentor
Ask Mentor: “Retest gate — minimum literacy?”
Knowledge Check
APPLY: AI red team on Cyberlium means:
Multiple choice
Knowledge Check
APPLY: True or False: Weaponized jailbreak kits are red team lab.
True or False
Knowledge Check
APPLY: Red team output should:
Multiple choice