Google blocked 100m spam Gmail messages using AI

TensorFlow helped detect the 0.1 percent of spam messages that slip past filters

Google's Gmail is used by 1.5bn people each month with 5m businesses using the service as part of G Suite and one of the biggest draws of the service is its built-in security protections.

Through the use of machine learning (ML), the company is able to block 99.9 percent of spam, phishing, and malware from ending up in user's inboxes.

However, by implementing new protections powered by Google's open-source machine learning framework TensorFlow, the search giant has succeeded at blocking an additional 100m spam messages each day.

• A look inside Google's security team
• How to make your email more secure
• Half of the malicious emails tied to credential phishing

Google is now blocking spam categories that were previously very hard to detect. By using TensorFlow to scan incoming emails, the company is now able to block image-based messages, emails with hidden embedded content and messages from new domains trying to hide a low volume of spam within legitimate traffic.

Using ML to block spam

ML is helping Google catch spam by allowing the company to identify patterns in large data sets that humans might not catch. The technology also makes it easier to adapt quickly to new tactics used by spammers while personalizing spam protections for each user as one person's spam may be an important message for someone else.

Applying ML at scale can be both complex and time consuming which is why TensorFlow contains many components to make ML easier and more efficient. TensorFlow Extended (TFX) is one of these components that allows Google to deploy ML pipelines in a quick and standardized fashion while TensorBoard allows it to monitor model training pipelines and quickly evaluate new models to determine their usefulness.

TensorFlow also provides the flexibility to train and experiment with different models in parallel. This helps businesses develop the most effective approach as opposed to being limited to running one experiment at a time.

Google is also experimenting with TensorFlow in other security-related areas such as phishing and malware detection in its ongoing goal to make the internet a safer place.



READ MORE: