Google launched security measures in early 2017 that were designed to combat spam and phishing attacks in Gmail using machine learning models. The Mountain View-based giant claimed at the time that the tools have since been able to detect unwanted and malicious email messages with 99.9% accuracy.
Now, Google has bolstered Gmail's anti-spam capabilities with TensorFlow, its open-source machine learning framework that now combines with its existing security filters for Gmail. The search giant says it's using that integration to ward off 100 million extra spam messages every day.
Neil Kumaran, Product Manager for Counter-Abuse Technology at Google, wrote in a blog post that TensorFlow helps the company expand its spam filters to additional types of unwanted emails that weren't previously easy to catch. These categories include messages that use images to hide malicious content and emails that take advantage of new domains to hide spam withinlegitimate traffic.
But Kumaran pointed out that Google is taking precautions so that it does not mistake important emails for spam. While applying machine learning to determine spam messages more accurately may take an extra amount of time, Kumaran said that TensorFlow works to speed up that process.
As the threat landscape continues to expand at a rapid rate, Google plans to apply TensorFlow to its fight against phishing and other forms of malware.