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 ProgressAI Security Research Programme
Ongoing
A vendor-neutral programme of hands-on AI security research, anchored on the OWASP Top 10 for LLM Applications, MITRE ATLAS and the NIST AI Risk Management Framework. OMEGA Core is the lab environment throughout. Research targets the platform's own systems only, and each item moves through one pipeline: question, map, design, execute, evidence, analyse, decide, publish and re-test.
Research Tracks
Six tracks. Each finding is published and becomes a regression test.
R1
Prompt Injection and Extraction
Direct and indirect prompt injection, system prompt leakage, and the measured effectiveness of prompt-level security controls.
R2
Model Behaviour and Alignment
Guardrails, aligned versus uncensored and abliterated models, guard-model classifiers and hallucination in security-focused role models.
R3
RAG and Data
Retrieval-augmented generation security: retrieval poisoning, embedding leakage and vector store access control.
R4
Agents and Tools
Excessive agency and tool misuse in agentic workflows, including the home-automation voice path.
R5
Platform and Supply Chain
Hosting security for AI workloads: endpoint exposure, container hardening, model and image provenance.
R6
Governance and Risk
Architecture baseline, risk register, model policy and AI incident response, mapped to NIST AI RMF.