Author Type

Graduate Student

Date of Award

Summer 5-19-2026

Document Type

Thesis

Publication Status

Version of Record

Submission Date

July 2026

Department

Information Technology and Operations Management

College Granting Degree

College of Business

Department Granting Degree

Information Technology and Operations Management

Degree Name

Master of Science (MS)

Thesis/Dissertation Advisor [Chair]

Nataliia Neshenko

Abstract

Interpreting cybersecurity incidents within critical infrastructure requires analysis across a complex information environment. Authoritative reporting, open web journalism, and Al-generated summaries coexist but often diverge. This study examines narrative properties across these three environments. Lexical uncertainty analysis, embedding-based semantic drift measurement, and probabilistic topic modeling are applied to 908 human-produced documents covering five U.S. critical infrastructure incidents. Results indicate that narrative divergence varies systematically by incident rather than by source type. Factors such as technical complexity, attribution ambiguity, and public visibility influence this divergence. Preliminary analysis of 429 Al-generated narratives indicates that large language models create a distinct environment, characterized by lower uncertainty, measurable divergence from authoritative baselines, and a thematic focus on public-facing content rather than technical detail. The source consulted first may shape an individual's preliminary understanding of the incident. This has potential implications for threat perception and attribution framing as AI systems become more widely used.

Included in

Business Commons

Share

COinS