A secure, end-to-end threat-intelligence pipeline turning raw LLM invocation logs into real-time, graph-aware defenses for Canadian security-operations centres.
This project aims to develop a secure, end-to-end threat-intelligence pipeline that transforms raw LLM invocation logs into real-time, graph-aware defenses for deployment in Canadian security-operations centres (SOCs). It unfolds across four main objectives: (1) secure ingestion and hybrid parsing of unstructured logs; (2) real-time anomaly detection and interactive dashboards; (3) development of an ontology of LLM-specific threats, and training of graph-based classifiers; and (4) system-wide evaluation using red-team self-play and deployment-ready continuous integration/continuous deployment (CI/CD) tools.
Elnaz Rabieinejad, Ali Dehghantanha, Fattane Zarrinkalam, J. Schwartzentruber
Applied Cryptography and Network Security Workshops (ACNS 2025) · 2025
Read paper