Reproduce “I can't recognize (yet): Delayed Rendering to Defeat Visual Phishing Detectors” on OpenResearch

About this paper

Ying Yuan, Cristiano Alex Rado, Giovanni Apruzzese, Mauro Conti, Luigi Vincenzo Mancini

Researchers from Sapienza University of Rome, University of Padua, Reykjavik University, and University of Liechtenstein demonstrated that state-of-the-art visual phishing detectors are vulnerable to timing-based, delayed-rendering attacks, achieving up to 100% evasion against common detection systems while simultaneously deceiving 72% of human users in a trust assessment.

Key insights

• Timing-based, delayed-rendering attacks represent a previously underexplored and highly effective adversarial technique capable of bypassing state-of-the-art visual phishing detectors.
• The effectiveness of rendering attacks against detectors is influenced by the attack type (curtain effect stronger than pixelation) and the detector's mechanism (logo-based detectors are more vulnerable to logo delays).
• Delayed rendering not only evades automated detectors but also largely deceives human users, who often fail to notice the manipulations and maintain high trust in adversarial webpages.

Summary by alphaXiv.