SecureMail AI: An Explainable Framework for Phishing Email Risk Assessment and Cybersecurity Awareness

This paper presents SecureMail AI, an educational email-security application developed to support the preliminary assessment of potentially malicious messages. The prototype accepts email-related information, including sender details, subject lines, and message bodies, and evaluates it against a collection of predefined indicators. These indicators include urgent requests, attempts to obtain credentials or sensitive information, financial language, attachment-related warnings, impersonation cues, and suspicious URL characteristics. The application combines detected indicators into a heuristic risk score ranging from 0 to 100 and assigns a corresponding risk category. It also presents explanations, extracts links for inspection, and supports downloadable analysis reports. Unlike a trained machine-learning classifier, the current implementation uses transparent rules rather than a model learned from a labeled dataset. This design makes the application easier to inspect, demonstrate, and modify, while also introducing limitations related to false positives, false negatives, and the inability to recognize sophisticated attacks that do not match its rules. The paper documents the system architecture, risk-assessment methodology, proposed dataset strategy, evaluation framework, and future research directions. It does not claim measured detection accuracy because a controlled evaluation has not yet been established. The central contribution is a reproducible educational prototype and an evaluation plan for investigating how transparent risk indicators can support phishing awareness and preliminary email triage.