AI-ASSISTED CYBERATTACK RISKS IN HIGHER EDUCATION: AN INTEGRATED DEFENSIVE FRAMEWORK AND SYNTHETIC PRIVACY EVALUATION

AI-assisted malware adaptation, phishing, and synthetic-media impersonation create interconnected security and privacy risks. This paper develops a design-oriented defensive framework for higher-education institutions, integrating privacy governance, identity assurance, behavioral monitoring, security operations, and independent human verification. Established guidance informs the architecture; a reproducible synthetic study evaluates only its privacy-processing components. Thirty independently seeded datasets of 10,000 fictional student records were generated. Experiments measured a quasi-identifier linkage-risk proxy, record suppression, aggregate count error under Laplace noise, and local batch-processing time. At k=5, the mean linkage-risk proxy among retained records was 0.02928, compared with 0.12207 for the unmodified synthetic microdata; mean record retention was 97.68%. For a separate department-count mechanism, mean absolute count error decreased from 10.45 at epsilon=0.1 to 0.45 at epsilon=2.0. These outcomes characterize the configured generator and processing mechanisms, not real-world re-identification probabilities or operational cybersecurity effectiveness. Malware detection, phishing resistance, deepfake verification, and end-to-end Zero Trust enforcement remain unevaluated. The contribution is an integrated architectural proposal supported by a transparent, limited synthetic demonstration.