In a Homoglyph Attack, also known as domain spoofing, an attacker tries to fool humans and computer systems by using such characters and symbols which may appear visually similar to characters used in the real domain but are different. Users are lured into clicking on the fake domain and redirected to a suspicious domain via which the attacker releases malware and collects sensitive information.
We at SecurityX use a cutting-edge solution known as “Siamese
Convolutional Neural Network (CNN).” to detect homoglyphs without
the need to provide paired data.
Instead of comparing strings to a standard list to detect the homoglyph, CNN uses a “learned” metric system. CNN algorithms are built to detect the visual similarity of the rendered strings. It does so by converting the domain names to feature vectors or images. These images are then indexed using randomized KD-Trees to perform a comparison. When a new domain name is observed, it is converted to an image and searched in the KD-Tree index to find any visually similar matches. If a match exists, then a homoglyph attack is detected. This technique shows a considerable 13% to 45% improvement over baseline techniques.
Our end-to-end anomaly detection system is useful in the real world to safeguard users’ critical information by preventing them from any such attack.
It detects anomalies across the business quickly and efficiently while reducing incident-related costs significantly. Built with a robust defense system, the SecurityX anomaly detection system works to handle any plausible threats at a lightning speed.
Using artificial intelligence (AI) along with machine learning (ML), we can help catch data abnormalities even before they impact your business.
Using AI models with ML is that they can be trained to automatically analyze datasets, decide what’s normal or abnormal behavior, and identify breaches in patterns quickly without any human intervention. With continuous self-learning mechanisms and the goldmine of massive knowledge it collects, the AI/ML model perpetually is in a state of predicting future anomalies rather precisely.
We developed state-of-the-art Machine Learning to predict whether there is an attack or not — in real-time. Our system was found to be accurate 99% of the time. For evaluation purposes, we compared our performance with other baselines and were found to outperform them.
© Copyrights 2021 Albot. All Rights Reserved.