Gmail from Google today has more than 1.5 billion users worldwide. Many people use their Gmail account on a daily basis and we also do that in many different languages. 15 years after the start of Gmail, spam is still not banned. Malware is often sent via a document that is attached to the e-mail.
AI tool to detect malware
The people and companies behind malware are getting smarter and spam cannot be completely filtered out by Gmail, but also Microsoft (Outlook). But Google is also making Gmail smarter to be able to filter as much spam as possible. A ‘deep learning’ AI (Artificial Intelligence) tool is now used to prevent malware from entering your inbox.
Google blocks almost all spam
Although you may find that you still receive a lot of spam in your mailbox, Google is able to filter out 99.9%. Google uses machine learning to improve the Gmail security process. In 2017, Google already announced that machine learning models 99.9% of all spam and phishing messages will not reach your inbox. That is of course a considerable achievement and in 2020 Gmail still manages to maintain that percentage. The machine learning models are constantly being refined.
To keep malware out as much as possible, the Gmail scanner processes 300 billion attachments every week in search of harmful documents. Of all documents that are blocked, Google says that 63% of them change daily. That is why the company is investing heavily in the next generation of scanners: AI Deep Learning tools.
For example, Google uses deep learning tools
You can see machine learning as a branch of AI that uses algorithms for which structured data must be entered in the system to work properly. To get this done, human intervention is needed. Deep learning is seen as a kind of human brain, a collection of all kinds of machine learning methods, based on so-called neural networks. Deep learning can very well categorize photos and translate talk messages. Google already uses deep learning for this and will now use it in the fight against malware.
Good malware scanning results
Since the end of 2019, Google has been working with a new deep learning scanner in the fight against malware. Since then, the daily detection coverage of Office documents containing malicious attachments has increased by 10%. That’s at 300 billion weekly attachments, a huge amount. Where the scanner mainly scores well is the detection of hostile ‘burst attacks’. Thanks to deep learning, the identification of these types of malicious documents has been improved by 150%.
It works with a TensorFlow model for deep learning and an adapted documentation analysis for every other file type. TensorFlow is an open-source software library used for data flow and differentiated programming, and Google is training its model with the TensorFlow Extended (TFX) platform.