Gmail’s Biggest Security Update: AI-Powered Spam Filter
The Google security blog has recently featured an overview of the most recent update made to the spam filters used by the Gmail email service provider. We’re talking about a new text classification system called Resilient & Efficient Text Vectorizer (RETVec). Google claims that the RETVec deployment is “one of the largest defense upgrades in recent years”.
As per Google developers, RETVec can now detect spam messages with “adversarial text manipulations” – emails containing typos, special characters, emojis, and having other characteristics of junk emails that were previously noticed by humans but were unrecognizable by machines.
Moreover, the new spam filtering algorithm effectively discerns email messages containing homoglyphs – characters that look similarly but have distinct meanings. Prior to the implementation of RETVec, such spam emails passed through Gmail’s barriers easily.
Its way-out algorithm allows RETVec to work out-of-the-box with every language and all UTF-8 characters. It can be used for mobiles and edge devices as well as for web applications.
The RETVec filtering system’s work is apparently similar to how a human reads emails. The algorithm uses the TensorFlow AI framework, which is looking at visual “similarity” to determine the meaning of words, rather than reading the written characters.
As the Google developers say, “replacing the Gmail spam classifier’s previous text vectorizer with RETVec allowed us to improve the spam detection rate over the baseline by 38% and reduce the false positive rate by 19.4%”.
The RETVec filtering system seems to be truly working as I am not seeing emails with spam elements in my Inbox anymore. Instead, they are going straight to my Gmail’s Spam folder.
Examples of spam emails detected by Google AI-Powered Spam Filter
Google presents RETVec as an open-source text vectorizer that allows anyone to build a reliable and efficient server-side and on-device text classifiers in order to block spam emails.