Workloads
Research Workloads
Workloads are the structured research tasks, labs and exercises running on OMEGA Core. Each workload produces documented findings — published in the Journal, reflected in the Research pages, and fed back into the platform as improved tooling, configuration or security controls.
Workloads are not tutorials. They are practical engineering investigations conducted on live infrastructure with real models, real data and real security implications.
Active Programme
In ProgressGIAC AI Platform Security (GAIPS) — 12-Week Lab Programme
July – September 2026
A structured twelve-week programme of hands-on security research aligned to the eight GAIPS knowledge domains. OMEGA Core is the practical lab environment throughout — all labs are conducted on the live platform, not against simulated or sandboxed systems. Findings are documented as they emerge.
Research Domains
Eight domains, twelve weeks. Each domain produces at least one documented lab finding.
01
AI and LLM Foundations
Capabilities, limitations and attack surfaces of generative AI systems. Model inventory, context windows, hallucination behaviour and prompt-level attack surface mapping.
02
AI Application Architecture
GenAI application security best practices, supply chain risk and integration-layer vulnerabilities. FastAPI surface assessment, model provenance audit.
03
AI Infrastructure Security
Security risks of hosting GenAI workloads. Container hardening, network exposure, least privilege across Docker stacks and Ollama LAN binding threat model.
04
AI Risk Management
Threat modelling methodologies and strategic risk planning. STRIDE/PASTA threat model for OMEGA Core. NIST AI RMF and EU AI Act risk classification applied to the platform.
05
Knowledge Augmentation and Retrieval
RAG architecture security, vector database risk and retrieval pipeline vulnerabilities. Indirect prompt injection via retrieval, Qdrant access controls, data poisoning evaluation.
06
Model Customisation and Alignment
Fine-tuning, moderation controls and alignment security. Modelfile system prompt evaluation, llama-guard3 safety classification, guardrail bypass research.
07
MLOps and Pipeline Security
Model lifecycle operations, data workflows and MLSecOps controls. Model pull-to-deploy pipeline audit, inventory API integrity, operator script security.
08
Agentic Systems and AI Integrations
Inter-agent communication, context management and protocol-level security. Tool use threat modelling, context injection, privilege escalation in agentic workflows.