AI-GENERATED DEEPFAKES AND CYBERSECURITY RISK PERCEPTION: INVESTIGATING ATTITUDES OF STUDENTS AT UNIVERSITY LEVEL

Mudassir Hussain; Tayyab Shahzad Akram; Muhammad Imran; Naveed Naeem Abbas; Dr. Muhammad Arfan Lodhi*

Authors

Mudassir Hussain
Tayyab Shahzad Akram
Muhammad Imran
Naveed Naeem Abbas
Dr. Muhammad Arfan Lodhi*

Abstract

The emergence of AI-based deepfakes has made synthetic media a cybersecurity threat, and deepfake attacks now include identity theft, phishing, CEO fraud and social engineering scams that have resulted in over USD 25 million in losses in one incident. Students of computing, as the next generation of cybersecurity professionals, are strategically positioned in a critical but empirically understudied role, as their deepfake awareness, perceptions and actions have implications for the cybersecurity of our country. Despite increasing research, there is limited empirical research on South Asian higher education students' perceptions of AI-driven deepfakes as cybersecurity threats, or on the theoretical links between deepfake awareness, attitude, cybersecurity risk perception and intention to engage in protective behavior. This research fills this gap with a quantitative, cross-sectional survey of 107 BSCS students in Pakistani higher education using a 30-item Likert-scale questionnaire and testing four hypotheses in a directional manner through Pearson correlation and linear regression in IBM SPSS, based on a serial mediation model that incorporates the Technology Acceptance Model and Protection Motivation Theory. All four hypotheses were confirmed at p < .01: deepfake awareness influenced attitude (β = .512), attitude influenced risk perception (β = .571), risk perception influenced protective behavior (β = .558) and attitude had the greatest direct effect on protective behavior (β = .603).Perceived personal risk was regarded as relatively moderate (M = 3.39) but perceived systemic danger was rated as substantial, indicating a theoretically important optimism bias. Students felt that universities provide inadequate training (M = 3.02), although they strongly supported government laws (M = 4.25) and curriculum inclusivity (M = 4.04).This research confirms a TAM-PMT mediation model for deepfakes and offers empirical insights for curriculum and institutional policy reforms, as well as national regulatory measures in the Global South.