Weaponising Fear: A Content Analysis of AI-Generated Panic and Escalation Claims on Social Media During the US–Israel–Iran Crisis

Abstract

Background: The weaponisation of digital platforms and generative artificial intelligence has transformed modern geopolitical conflict, turning social media into strategic channels for synthetic escalation narratives. Despite growing concern regarding synthetic media during international crises, empirical investigations into how generative artificial intelligence is operationalised to manufacture public panic and threat perceptions during acute military confrontations remain scarce.

Objective: Grounded in Entman’s Framing Theory and the Information Disorder conceptual framework, this study investigated the manifest modalities, threat frames, rhetorical devices, engagement dynamics, and platform moderation interventions associated with AI-generated panic claims during the US–Israel–Iran standoff.

Methodology: A quantitative content analysis was conducted on a purposive sample of verified AI-generated social media posts (N = 420) retrieved from X (n = 268) and Telegram public channels (n = 152) across critical crisis escalation windows. Manifest and latent variables—including synthetic modality, threat frames, credibility cues, and platform flags—were systematically coded following established inter-coder reliability protocols. Descriptive statistics, one-way ANOVA with post-hoc Tukey tests, independent-samples t-tests, and Chi-square tests of independence were utilised for empirical analysis.

Results: Photorealistic static imagery emerged as the dominant synthetic modality (51.90 per cent), followed by deepfake videos (27.14 per cent). Escalation framing was heavily dominated by immediate kinetic attack claims (44.29 per cent) and global contagion or World War III narratives (29.05 per cent). Posts were heavily saturated with urgent panic markers (80.48 per cent) and spoofed legacy broadcast authority cues (39.05 per cent), with the latter significantly accelerating dissemination velocity (p < .001). High-severity non-conventional or nuclear escalation frames generated significantly higher public engagement than non-kinetic or economic frames (p < .001). Furthermore, 71.67 per cent of the synthetic content remained entirely unmoderated, with dynamic multimodal deepfakes evading platform-level automated detection at rates exceeding 91 per cent.

Conclusion: Generative artificial intelligence operates as a potent instrument of crisis induction and cognitive destabilisation during militarised standoffs. Commercial platforms demonstrate systemic latency in moderating dynamic multimodal synthetic artefacts, leaving digital information spaces vulnerable to calculated, fear-driven manipulation during international conflicts.