EventsThe 1st International Online Conference on Forecasting
Published
This submission belongs to the session S2. AI Forecasting & Large Language Models of the event The 1st International Online Conference on Forecasting
Published date
16 Sep, 2026
Academic Editor
author-avatarSonia Leva
Citation
Fahim Sufi, Forecast-Ready Cyber Threat Salience Signals from Global Cybersecurity News Metadata, in Proceedings of The 1st International Online Conference on Forecasting, 21 September–22 September 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

Forecast-Ready Cyber Threat Salience Signals from Global Cybersecurity News Metadata

image
1. Office of CTO, COEUS Institute, New Market, VA 22844, USA
2. CTI, University of Dhaka, Dhaka 1000, Bangladesh
Abstract

Introduction:
Cybersecurity forecasting requires timely signals that reveal how public threat visibility changes before these signals are formalised in incident repositories, vulnerability databases, or institutional risk assessments. This study develops a forecast-ready cyber threat salience framework that transforms global cybersecurity news metadata into temporal, thematic, and source-aware indicators for cyber risk monitoring.

Methods:
The study analyses 18,307 usable cybersecurity news records collected between 3 October 2023 and 2 June 2026. Temporal analysis uses 18,263 complete month records across 32 months, derived from 190 source domains, 18,248 unique title strings, and 17,922 unique URLs. The framework combines title level multi label thematic coding, monthly aggregation, early to late salience comparison, Jensen–Shannon divergence, and source concentration metrics. Eleven cyber threat themes were measured, including vulnerabilities and exploits, malware and botnets, AI security, nation state activity, phishing, data breaches, ransomware, cloud and identity, critical infrastructure, supply chain compromise, and cryptocurrency-related risks.

Results:
Cybersecurity reporting behaved as an episodic forecasting signal, with a 3.4-fold difference between the highest and lowest complete month volumes. The most visible theme was vulnerabilities and exploits, appearing in 2,816 records, or 15.4% of usable records, followed by malware and botnets with 1,509 records and AI security with 1,407 records. Late period salience increased most strongly for vulnerabilities and exploits, from 13.5% to 21.3%, and AI security, from 6.0% to 13.6%, while ransomware declined from 5.8% to 1.6%. Source concentration was high, with the top five domains contributing 60.9% of all records.

Conclusions:
The framework supports cyber forecasting by converting public cybersecurity reporting into measurable salience signals for horizon scanning, AI security monitoring, and source-aware cyber risk intelligence.

Keywords
Cyber Threat Forecasting
Cyber Threat Salience
Cybersecurity News Metadata
AI Security Visibility
Temporal Risk Signals
Source Concentration
A Receding-Horizon Decision-Focused Learning Framework for Cost-Optimal Battery Scheduling in Microgrids
Predicting Next Moves: Customer Nowcasting Using Data Science