OWASP LLM Top10
OWASP LLM Top 10 literacy — prompt injection, insecure output, supply chain, excessive agency — on YOUR `$AI_LAB` app stubs only.
Visual · t32_owasp_llm_top10
OWASP LLM Top10 = named risk rows. $AI_LAB. Original Cyberlium.
LLM apps ship new attack surfaces — name OWASP LLM Top 10 rows on YOUR lab chatbot before trusting prod guardrails alone.
OWASP LLM Top 10 literacy names: LLM01 prompt injection, LLM02 insecure output handling, LLM03 training data poisoning, LLM04 model denial of service category, LLM05 supply chain vulnerabilities, LLM06 sensitive information disclosure, LLM07 insecure plugin design, LLM08 excessive agency, LLM09 overreliance, LLM10 model theft category. Analyst maps five Top 10 rows to `$AI_LAB` instructor chatbot stub or YOUR toy app — documents defender control per row — without jailbreak weaponization against stranger apps, without publishing bypass kits, without testing employer prod LLM without RoE. Cyberlium teaches named literacy on labeled lab apps — NOT how to weaponize injection against unauthorized targets. Refused: stranger app jailbreak kits, prod API abuse without ticket, malware-gen prompts. Lab row: Top 10 mapping table (risk ID, lab surface, defense category) five rows. chmod 600 under $HOME/cyberlium-lab/t32-ai/.
1. Named Top 10 rows
LLM01–LLM05 anchor prompt injection, output handling, poisoning, DoS, supply chain — five starter literacy rows.
Each row maps to observable lab app behavior — not stranger SaaS exploitation.
Command guide
Try these commands — Named Top 10 rows
═══ LINUX / macOS (Terminal Practice) ═══
Check system state and user context
Command — copy this
id whoami uname -a
Inspect network sockets listening for connections
Command — copy this
ss -tuln 2>/dev/null || netstat -tuln
Audit active processes
Command — copy this
ps aux | grep -v "\[" | head -15
═══ WINDOWS (POWERSHELL) ═══ Query user identity and system information
Command — copy this
whoami /all Get-ComputerInfo | Select-Object CsName, OsName, OsVersion
Primary tools to practice this lesson: curl, grep. Reference sites: OWASP LLM Top 10 (https://owasp.org/www-project-top-10-for-large-language-model-applications/); OWASP LLM cheatsheet (https://cheatsheetseries.owasp.org/cheatsheets/LLM_Top_10_Cheat_Sheet.html); MITRE ATLAS (https://atlas.mitre.org/). Run every command in the box — install first, then the usage lines — only on YOUR lab / program scope.
2. Defense mapping
Input validation, output encoding, least agency, secrets hygiene, eval gates — link prior modules.
Document which control mitigates which Top 10 row on YOUR `$AI_LAB` stub.
3. $AI_LAB boundary
YOUR toy app or instructor stub only — not stranger chatbot APIs.
Refused: jailbreak weaponization kits, prod LLM abuse without RoE, malware-gen prompt cookbooks.
4. What you ship: OWASP LLM Top 10 mapping table
Five Top 10 rows + defense category each + NEVER jailbreak weaponization line.
5. What you record before the next lesson
OWASP LLM Top 10 mapping table path.
6. Wrong vs right: stranger SaaS vs YOUR toy LLM
Worked failure — same MSF word, opposite target. Right never needs a café Wi-Fi or classmate laptop.
- Wrong: Publish jailbreak kit against public ChatGPT clone as 'Top 10 research.'
- Right: Top 10 mapping from `$AI_LAB` toy app stub. Next: RAG and Plugins.
Mission: OWASP LLM Top 10 mapping table
1) Name five OWASP LLM Top 10 risk IDs. 2) Map each to lab app surface. 3) Defense category per row UTC. 4) Write NEVER jailbreak weaponization line.
Stuck? Ask Cyberlium AI Mentor
Ask Mentor: “LLM08 excessive agency — literacy default?”
APPLY: OWASP LLM Top 10 on Cyberlium uses:
APPLY: True or False: Weaponized jailbreak kits are course lab.
APPLY: Top 10 literacy includes: