Investigating system intrusions with data provenance analytics

To aid threat detection and investigation, enterprises are increasingly relying on commercially available security solutions, such as Intrusion Detection Systems (IDS) and Endpoint Detection and Response (EDR) tools. These security solutions first collect and analyze audit logs throughout the enterprise and then generate threat alerts when suspicious activities occur. Later, security analysts investigate those threat alerts to separate false alarms from true attacks by extracting contextual history from the audit logs, i.e., the trail of events that caused the threat alert. Unfortunately, investigating threats in enterprises is a notoriously difficult task, even for expert analysts, due to two main challenges. First, existing enterprise security solutions are optimized to miss as few threats as possible – as a result, they generate an overwhelming volume of false alerts, creating a backlog of investigation tasks. Second, modern computing systems are operationally complex that produce an enormous volume of audit logs per day, making it difficult to correlate events for threats that span across multiple processes, applications, and hosts. In this dissertation, I propose leveraging data provenance analytics to address the challenges mentioned above. I present five provenance-based techniques that enable system defenders to effectively and efficiently investigate malicious behaviors in enterprise settings. First, I present NoDoze, an alert triage system that automatically prioritizes generated alerts based on their anomalous contextual history. Following that, RapSheet brings benefits of data provenance to commercial EDR tools and provides compact visualization of multi-stage attacks to system defenders. Swift then realized a provenance graph database that generates contextual history around generated alerts in real-time even when analyzing audit logs containing tens of millions of events. Finally, OmegaLog and Zeek Agent introduced the vision of universal provenance analysis, which unifies all forensically relevant provenance information on the system regardless of their layer of origin, improving investigation capabilities.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Hassan, Wajih Ul

Contributors dc:contributor

• Bates, Adam
• Bailey, Michael
• Gunter, Carl
• Paxson, Vern
• Xu, Dongyan

• Computer science

Statement dc:rights

• Copyright 2021 Wajih Ul Hassan

Language dc:language

en, eng

Handle dc:identifier

http://hdl.handle.net/2142/113904

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F113904