This research introduces PhishOracle, a novel tool for evaluating the robustness of phishing detection models against adversarial attacks.

• Reveals vulnerabilities in existing ML and deep learning phishing detection systems
• Demonstrates how attackers can bypass detection by manipulating webpage elements
• Tests detection models against LLM-generated adversarial phishing pages
• Provides a framework for security teams to assess and improve phishing defenses

As phishing attacks grow increasingly sophisticated, this research provides critical insights for cybersecurity professionals to strengthen detection systems against evolving threats.

From ML to LLM: Evaluating the Robustness of Phishing Webpage Detection Models against Adversarial Attacks