AI-Identity-Defense-ITDR: A Hybrid Deterministic and Unsupervised Machine Learning Framework with Explainable AI for Active Directory Identity Threat Detection and Response
Enterprise Active Directory (AD) environments remain the primary target for post-exploitation privilege escalation and lateral movement, with credential harvesting attacks such as Kerberoasting frequently evading traditional signature-based Security Information and Event Management (SIEM) thresholds. Simultaneously, integrating cloud-hosted Large Language Models (LLMs) into Security Operations Center (SOC) workflows introduces severe data-sovereignty and telemetry-leakage risks when processing sensitive directory logs. This paper presents AI-Identity-Defense-ITDR (v2.0), an open-source, privacy-preserving Identity Threat Detection and Response (ITDR) framework designed to detect, visualize, and explain Active Directory identity attacks locally. The architecture combines raw Windows Event Log (EVTX) telemetry parsing with a dual-layer detection engine: deterministic behavioral rules targeting Kerberoasting indicators (Event ID 4769 RC4-HMAC encryption downgrades) and an unsupervised Isolation Forest User and Entity Behavior Analytics (UEBA) model built with scikit-learn, alongside interactive PyVis blast-radius attack graphs, side-by-side Explainable AI (XAI) feature attribution views, cryptographic AES-256 remediation simulation, and local SecOps briefings powered by Ollama (Llama 3).
Authors
- Kavana Arabagatta Shivarudrappa
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24382543
- Primary Topic
- Network Security and Intrusion Detection
- Type
- preprint