Preventative Measures for Online Scams
by A.J. Kleber
We’ve all seen the warnings and the news stories about defrauded victims; the internet can be a dangerous place, full of scammers hoping to take advantage of ever-more-sophisticated camouflage and large-scale automation to steal from the unwary. Indeed, e-commerce scams are annually responsible for millions of dollars in financial damages across the globe, with recovery of funds proving difficult to impossible in most cases. To counter this threat, systems have been developed to detect and assess fraudulent websites; however, while these systems are highly accurate when provided with websites to analyze, locating potentially malicious sites is challenging. Ideally, security measures would proactively identify scams and prevent harm, but relying on user reports is inherently reactive and often slow. The influx of new scam types is also extremely difficult to keep up with, and search engine queries utilized by protective systems often fail to generalize to evolving criminal behavior. It’s a concerning prospect for the average user.
Fortunately, there are researchers like Professor Gianluca Stringhini and Pujan Paudel (ECE PhD ‘25) on the case. In February, the pair received a Distinguished Paper Award at the 2026 Network and Distributed Systems Security Symposium (NDSS) for an effective new system designed for improved discovery of fraudulent websites, including those in newly-emerged scamming categories.
New strategies to combat fraud
LOKI, Stringhini and Paudel’s new system, uses a combination of strategies: a keyword scoring model grounded in the machine learning paradigm Learning Under Privileged Information (LUPI), and a machine learning training method known as “model distillation.” In the paper, they demonstrate an over-20x improvement in scam discovery, and using keywords identified using a small set of 1,663 scam websites, were able to detect 52,493 previously unreported fraudulent sites. Furthermore, LOKI demonstrates a capacity to generalize to brand new categories of scam, making it highly effective for detecting emerging threats. It’s a promising tool to help make traversing the online world safer for everyone.
Professor Gianluca Stringhini’s research is centered on system security and combating a wide variety of digital harm, from cyberbullying to malware to internet-based scams. His work has been supported by the NSF, Google, and a Red Hat Collaboratory Research Incubation Award. His accolades include early career awards from the NSF and the BU College of Engineering, and several Best/Distinguished Paper Awards. His research has been featured by the BBC, the New York Times, and other major news outlets.
Dr. Pujan Paudel is a Senior Data Scientist at Analysis Group. He received his doctorate in Computer Engineering at Boston University in Fall 2025, followed by 5 months of post-doctoral research in LLMs & AI at Truveta.