"Researchers Create New Approach to Detect Brand Impersonation"

Security researchers at Microsoft created a new method for detecting brand impersonation attacks. These attacks refer to the crafting of content to mimic a trusted company or known brand to trick unsuspecting victims into responding and disclosing information. Brand impersonation attacks have become increasingly hard to detect due to the continued advancement of technology and techniques. Justin Grana, an applied researcher at Microsoft, says these attacks have increased in accuracy from a visual perspective as brand impersonation can look the same as legitimate content. Today's brand impersonation attacks lack copy-and-paste or jagged logos, making it more difficult for people and technology to pick up on visual cues that previously helped distinguish fake content from true content. To address the challenge of detecting brand impersonation attacks, the team developed and trained a Siamese Neural Network on labeled images. Siamese Neural Networks are designed to make better predictions using a smaller number of samples, unlike standard deep learning, which is trained using many examples. The team's dataset contains over 50,000 screenshots of malicious login pages covering more than 1,000 brand impersonations. Each image is a collection of numbers, which were translated into what is described as a point on an N-dimensional coordinate plane. The team tried to make the numbers meaningful in order to distinguish fake from real brand images. The algorithm used by the team was rewarded for translating content of the same brand to similar numbers and content of different brands to different numbers. Any numbers observed to be close together were likely from the same brand. This article continues to discuss the concept of brand impersonation attacks, the Siamese Neural Network developed to detect these attacks, and lessons learned from the research behind this approach. 

Dark Reading reports "Researchers Create New Approach to Detect Brand Impersonation"