CPC H04L 63/1483 (2013.01) [G06N 3/08 (2013.01)] | 30 Claims |
30. A non-transitory computer-readable storage medium including instructions that when executed on a processor perform a method for detecting phishing attacks, the method comprising:
acquiring a set of legitimate web content from multiple webpages hosted at a plurality of legitimate top-level domains (TLDs);
generating, using an HTML5-compatible background browser, a set of legitimate browser images from the set of legitimate web content, wherein the legitimate browser images are bit-map images stored in memory without displaying content on an interactive display;
determining, using a neural network, a set of legitimate visual characteristics from the set of legitimate browser images;
accessing a hyperlink embedded within an email message in a user's email inbox;
generating, using the HTML5-compatible background browser, a background browser image from unauthenticated web content associated with the hyperlink;
identifying, using the neural network, a statistical similarity between visual characteristics of the background browser image and the set of legitimate visual characteristics;
determining that a domain of the unauthenticated web content does not match any of the plurality of legitimate TLDs associated with the statistically similar legitimate visual characteristics; and
responsively recording that the unauthenticated web content is illegitimate.
|