Hands-On AI Security
A 4-part practical laboratory on local LLM deployment, red-team prompt injection, Flask gateways, and OPA policy enforcement.
Key Learning Outcomes & Architecture Capabilities
- Deploy a zero-cost local LLM security testing laboratory using Docker Compose and Ollama
- Execute 5 distinct red-team prompt injection attacks against an intentionally vulnerable LLM ('Piper')
- Construct a hardened Python Flask reverse proxy security gateway with input/output guardrails
- Implement Policy-as-Code governance using Open Policy Agent (OPA) and declarative Rego rules
🗺️ Series Roadmap & Part Execution
A sequential, part-by-part engineering roadmap translating architecture into code. Published parts are live and open access; upcoming parts follow our weekly publication runway.
Hands-On AI Security: Building a Zero-Cost Local LLM Lab with Docker, Ollama, and Piper
Setting up a 100% free, local AI security sandbox using Docker Desktop, Ollama (llama3.2), and a custom vulnerable chatbot ('Piper').
Hands-On AI Security: Red Teaming Piper with 5 Prompt Injection Vectors
Probing and breaking local LLMs using Direct Injection, Roleplay, Obfuscation, and Authority Claims.
Hands-On AI Security: Blue Teaming and Building the Security Gateway
Decoupling AI defense from the model by building a Python Flask security proxy with hot-reloading ingress filters and DLP rules.
Hands-On AI Security: Policy-as-Code with Open Policy Agent and Rego
Moving beyond static regex filters by using Open Policy Agent (OPA) and Rego to enforce declarative, explainable security policies.
🏛️ Academic Research Foundation & Zenodo DOIs
This operational series synthesizes formal cybersecurity research conducted at the University of San Diego. The underlying theoretical frameworks, cryptographic proofs, and architectural blueprints are published with permanent Digital Object Identifiers (DOIs) and full PDF downloads:
Automating AI Security Risk Assessment, LLM Governance, and Threat Detection
Academic framework on automated AI risk classification, OWASP LLM mitigation pipelines, and runtime prompt auditing.
Open-Source Laboratory Code & Sandbox
Open-source Docker Compose sandbox, vulnerable banking chatbot ('Piper'), security gateway, and student attack worksheets.