Gmail is the largest email provider in the world with over 1.8 billion users. The service is top-rated due to its free nature and ease of integration with other Google services. However, this vast user base makes Gmail the favourite target for spammers and even scammers. In its cat-and-mouse chase with spam, Google has now implemented an AI-powered spam detection system called RETVec, which the company calls its biggest security upgrade in years. Given how AI has become the buzzword for 2023, it is not surprising to see Google implement this to help with security of its popular products.
Here’s everything you should know about Google’s RETVec that is capable of reducing spam on Gmail by 38%.
How RETVec Solves Gmail’s Challenges in Spam Detection
Although Gmail has been actively updating its algorithms and filters to prevent spam and phishing emails, tricksters have figured out ways to escape getting detected. Till now, Gmail has been using a standard UTF (Unicode Transformation Format) based text detection where it would filter out certain words and phrases from a pre-defined library or words.
For example, when Gmail tries to detect and filter a phrase like – ‘Congratulations, you have won a free trip’, it will try to find words and phrases that are similar to this statement. However, spammers use homoglyphs to escape this detection, by using characters that look similar to text alphabets but are not really alphabets.
In this case, if the scammer replaces ‘Congratulations’ with ‘CongratuIations’. It is a very subtle change for the human eye to notice, but Gmail will not consider it as spam, as it does not match with its spam dictionary. Given the millions of possible homoglyphs for a single word, it becomes very difficult for Gmail to control spam.
To tackle this, Gmail will now introduce RETVec which stands for Resilient and Efficient Text Vectorizer. It is an AI-powered algorithm that is trained to detect all types of homoglyphs and possible substitutions that a spammer can use to escape Gmail’s spam detection.
Google says that RETVec can detect spam in over 100 languages. It can also filter out spam based on special characters, especially with symbols that are not encoded with the standard UTF-8. For example, ‘Hello’ can also be written as ‘ℌ𝔢𝔩𝔩𝔬’ which looks fancy, but uses a non-standard text encoding which cannot be detected by Gmail’s normal spam filters.
RETVec will be an open-source algorithm. Gmail also says that it will be less resource-intensive, despite being an AI model. As per Google’s Security Blog, RETVec requires so little computing power that it can function even locally on a smartphone.
As per Gmail’s initial testing, RETVec will be able to reduce spam by over 38%. It will also reduce the resource utilization at Google’s servers by up to 78% which was previously high due to dictionary-based spam filtration. RETVec will also reduce the false positive rate of spam email by almost 20%.
Google says that RETVec has been in development and testing since last year. The company has already started deploying its latest AI-powered spam filtration for all Gmail users. Google is also calling RETVec its largest security upgrade in years.
