Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Watch this 12-minute conference presentation from SOUPS 2025 that introduces SmishX, an innovative SMS phishing detection system that uses large language models to not only identify malicious text messages but also provide clear explanations to help users understand why a message is dangerous. Learn how researchers from the University of Illinois Urbana-Champaign and OSF Healthcare addressed the critical challenge of SMS phishing attacks, particularly those targeting older adults, by developing a system that gathers external context such as domain information, brand verification, URL redirection analysis, and web screenshots to enhance the reasoning capabilities of LLMs. Discover how SmishX employs chain-of-thought reasoning to analyze short SMS messages that typically lack sufficient context for security assessment, then converts this analysis into user-friendly explanations that significantly improve phishing detection across all age groups. Explore the evaluation results showing SmishX achieves 98.8% accuracy while outperforming existing detection methods, and examine findings from user studies with 175 participants that demonstrate the system's "excellent" usability rating and its effectiveness in helping users make better security decisions. Gain insights into the ongoing challenges of resolving human-AI disagreements in cybersecurity contexts and the complexities of safely handling AI errors in real-world phishing detection scenarios.
