AI Security Visibility in Cybersecurity News Metadata: Measuring the Rise of Deepfake, LLM, and Adversarial AI Risk Salience

Introduction:
Artificial intelligence is rapidly transforming imaging, computer vision, biometrics, deepfake generation, and visual decision systems. These advances also create new cyber risks, including adversarial manipulation, synthetic media abuse, identity compromise, and AI-enabled social engineering. This study develops a reproducible cyber threat salience framework for measuring how AI security and related cyber risks become visible in global cybersecurity news metadata.

Methods:
The study analyses 18,307 usable cybersecurity news records collected between 3 October 2023 and 2 June 2026. Complete month analysis uses 18,263 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 analysis. Eleven cyber threat themes were measured, including AI security, vulnerabilities and exploits, malware and botnets, 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 visibility signal, with a 3.4-fold difference between the highest and lowest complete month volumes. AI security appeared in 1,407 records, representing 7.7% of all usable records. Its salience increased sharply from 6.0% in the early period to 13.6% in the late period, a gain of 7.6 percentage points. Vulnerabilities and exploits also increased from 13.5% to 21.3%, while ransomware declined from 5.8% to 1.6%. Source concentration was substantial, with the top five domains contributing 60.9% of records.

Conclusions:
The study shows how cybersecurity news metadata can quantify emerging AI security visibility. The framework supports source-aware monitoring of deepfake, LLM, adversarial AI, and cyber risk narratives relevant to future imaging and intelligent visual systems.