Since his arrival in 2004, Gmail has turned e-mail upside down. According to Google, the service attracted 1.5 billion users. While much has changed over the past 15 years, many aspects of Gmail have also remained unchanged, such as the invasion of spam in boxes.
One of the priorities of Google and its messaging service is to combat malware, especially those that pass through attachments received in your mailboxes. Macroviruses, which mainly infect Microsoft Word documents, existed long before Gmail was created. As early as 1995, the WM virus. Concept was the first to touch The Office Pack documents, and since microsoft has been selecting macro deactivation in Office by default. However, the problem has not disappeared. Malware transferred to attachments has multiplied, as have protection solutions. Google estimates that malicious documents currently account for 58% of all malware that targets Gmail users. Today, Google is fighting back with a deep learning AI to prevent this virus from reaching your inbox.
Google blocks 99.9% of malicious attachments in Gmail
Unsurprisingly, we learn of Google’s decision to invest in security. Earlier this year, the group had paid hackers $6.5 million to secure the internet, and had also presented its preventive measure to suspend all paid extensions of the Chrome online store as soon as an increase in fraud is detected. So it’s only natural that Google should turn to deep learning systems to secure Gmail. In 2017, Google announced that this type of system could block 99.9% of spam and phishing messages. At the time, this figure was impressive given that more than 50% of all messages received by Gmail were spam at the time.
Today, by 2020, deep learning has been perfected, and this 99.9% success rate is still true for spam, phishing and malware blocking. But it is the search for malware that has evolved. The Gmail scanner processes some 300 billion attachments each week, looking for malicious documents to block. Of the blocked documents, Google says 63% of them are different. It is this ever-evolving threat from malicious documents that has prompted Google to deploy a new generation of machine learning scanners: deep learning.
How Google uses deep learning to protect your inbox
Deep learning is already used by many companies, and many readings will allow you to learn a little more about how it works. At the risk of simplifying the concept to the extreme, machine learning can be seen as a branch of AI that uses self-modified algorithms that need structured data built into the system to function properly. In other words, the system needs human intervention. Deep learning is more like the human brain, in that it uses a neural network approach to data processing. The aim is to stack layers of these networks of artificial neurons on each other, to create a “deep” neural network. Deep learning is very useful for certain tasks, such as identifying photos and ranking them according to categories, or understanding voice commands. In fact, Google is already using this technology for some of its services.
Deep learning detection rates are on the rise
According to Google, the new scanner that works with deep learning has been active since the end of 2019. During this period, it increased the “daily detection of Office documents containing malicious scripts” by 10%. Given the amount of documents scanned by Google each day, this result is impressive. It is even more so in terms of “detecting enemy and burst attacks”, i.e. the type of massive distribution of documents carried out by botnets (a network of computer bots), which tend to be done on an ad hoc basis rather than at a steady pace. In these cases, deep learning improved the identification rate of malicious documents by 150%.
Jake Moore, cybersecurity specialist at ESET, says: “Malware is evolving at a pace that the security industry is struggling to keep up with. However, the use of deep learning seems to be able to minimize the risk of malware reaching inboxes around the world.”