SafeLink: Toward Inclusive Mobile Security Through Real-Time Phishing Prevention

초록

Phishing remains a persistent and rapidly evolving threat, particularly in mobile-first ecosystems where users interact with unverified links via QR codes and SMS. Traditional blacklist filters and static scanners are ineffective against zero-day or obfuscated attacks, leaving vulnerable groups - such as the elderly - at heightened risk. We present SafeLink, a mobile browser framework that blocks malicious content before rendering by integrating URL-level machine learning with LLM-based semantic analysis of normalized HTML. Evaluations on the TR-OP benchmark show that SafeLink outperforms representative baselines with an F1 score of 0.967, while structural HTML preprocessing reduces token usage by 89.6%. © 2026 Copyright held by the owner/author(s).

키워드