🧪 Challenge Track: Applied AI Security
Vendor-neutral. Free. Practical. Five labs that turn your cybersecurity experience into portfolio evidence for AI Security and Governance roles — without depending on any vendor.
AI security job postings (AI Security Engineer, Detection Engineer, AI Security Specialist, Security Consultant) ask for one thing above all: having done it, not only studied it. Each challenge here produces a verifiable artifact (a report, a threat mapping, an assessment) that you can publish on GitHub and explain in an interview. All tools are open source and free.
🎯 Who it is for
IT / cybersecurity / forensics professionals repositioning toward AI. You do not need to be a senior developer: you need basic Python and discipline. It complements Phase 3: Cybersecurity and Phase 5: Cloud + AI of the plan.
🧰 The 5 challenges
| # | Challenge | Open tool | Frameworks | Portfolio artifact |
|---|---|---|---|---|
| 1 | LLM audit (OWASP Top 10) | garak | OWASP LLM Top 10 | Vulnerability report for an LLM |
| 2 | AI red-teaming + threat mapping | PyRIT | MITRE ATLAS | Red-team report with mapped TTPs |
| 3 | Governance assessment | (none — document) | NIST AI RMF + ISO/IEC 42001 | Executive gap assessment |
| 4 | PII detection/anonymization (DLP) | Presidio | GDPR / LFPDPPP | Data classification pipeline |
| 5 | Secure code review of AI-generated code | Semgrep + Gitleaks | OWASP Top 10:2025 + CWE Top 25 | SAST + secrets findings, mapped and fixed |
🧭 How these challenges help you get hired
Each challenge is anchored to real responsibilities in current AI security job descriptions:
| Typical responsibility in JDs | Challenge that evidences it |
|---|---|
| "Interpret AI alerts, reduce false positives, tune detection" | Challenge 1 + Challenge 2 |
| "AI red-teaming / threat hunting; adversarial threat intelligence" | Challenge 2 |
| "AI security requirements, compliance (EU AI Act, NIST AI RMF), governance and ethics" | Challenge 3 |
| "Identification and classification of sensitive data (DLP, PII), data protection" | Challenge 4 |
| "Review, triage, and remediate AI-generated code (SQLi, XSS, path traversal, auth bypass, secrets, SSRF)" | Challenge 5 |
A challenge is not "finished" until it meets the capstone rubric: professional README, reproducible steps, diagram, decisions/trade-offs, and a 5-minute demo you can explain without notes.
⚙️ Step 0 — Isolated environment (one time, ~5 min)
Challenges 1, 2, and 4 use Python. Challenge 5 also uses Python (for Semgrep) plus the Gitleaks binary. Create one reusable environment for all of them.
Where to run this: on your own machine (Windows, macOS, or Linux). You do not need cloud or a credit card.
# Crea y entra en la carpeta del track
mkdir ai-security-labs && cd ai-security-labs
# Entorno virtual de Python (requiere Python 3.10–3.12)
python -m venv .venv
# Windows (PowerShell): .venv\Scripts\Activate.ps1
# macOS/Linux: source .venv/bin/activate
# Carpeta local para tus reportes (tu "evidencia")
mkdir reports
Which model/LLM should I use for testing? (free options)
You do not need to pay for a model to practice. Options:
- Local model with Ollama (free, runs on your machine):
ollama run llama3.2and point the tools to the local endpoint. Recommended for practicing at no cost and with low risk. - Free tier from a provider (OpenAI, Google, Anthropic, Azure) if you already have access — use a key with a spending limit.
- Hugging Face model downloaded locally.
⚠️ Only test models you own or have explicit permission to test. Red-teaming a third-party system without authorization is illegal.
✅ Track checklist
- Isolated environment created (Step 0)
- Challenge 1: garak report published on GitHub
- Challenge 2: red-team report mapped to MITRE ATLAS
- Challenge 3: NIST AI RMF + ISO 42001 gap assessment
- Challenge 4: PII detection pipeline with Presidio
- Challenge 5: AI-generated code vulnerabilities detected, mapped (CWE/OWASP), and fixed
- All 5 artifacts meet the capstone rubric
- LinkedIn and CV updated with these projects
🔗 Resume Value
"I executed LLM security audits (OWASP LLM Top 10) with garak, AI red-teaming mapped to MITRE ATLAS with PyRIT, a governance gap assessment aligned to NIST AI RMF and ISO/IEC 42001, a PII detection pipeline with Presidio, and secure code reviews of AI-generated code (SQL Injection, XSS, Path Traversal, Auth Bypass, Secrets in Code, SSRF) with Semgrep and Gitleaks mapped to OWASP Top 10 and CWE — all published as reproducible evidence on GitHub."
All tools and frameworks in this track are verified in Sources and Verification.