Key Takeaways
- Google Messages offers spam protection for SMS, and the feature is enabled by default.
- The documentation for the feature has been modified to clarify that on-device machine learning detects spam patterns, which is then used to train AI models.
- AI models are used where on-device ML isn’t possible. Moreover, encrypted messages are spared from spam detection training.
Android is an evolved operating system and it offers several inbuilt protections against dangers like unauthorized app installation, but Google takes it a step further with everyday apps like Messages and the dialer, pre-installed on Pixel phones and several other devices. One such feature in Messages is called Spam detection, and as the name suggests, it helps ward off disturbing and dangerous SMS. Google has documentation describing how it works, but we just noticed the company makes a few changes there to add specificity. Unsurprisingly, AI is involved.
Google Messages has spam protection turned on by default, so you will see an alert message upon opening texts from a suspicious sender. Ideally, the feature tries to prevent delivery and redirects such unencrypted communication to the app’s Spam folder. This distinguishes scams from genuine messages, and Google documentation explains how this works, previously titled “Your chats stay private with spam detection,” the page is now called “How Google protects your privacy with spam detection.” We aren’t sure if the way this feature works has changed, but the new documentation spells out the role of AI in clearer terms.
First, Google uses on-device machine learning to see patterns in your messages and detect spam. Now, it says it also checks for URLs in the inbound messages, and contained URLs are uploaded to Google to ensure the link isn’t malicious. Additionally, the description now includes a big section about how spam detection may process suspicious texts and “signals” from them may be used to enhance AI models for spam detection. Why, you ask? Because the training data you supply enables the spam protections where on-device processing isn’t possible.
Encrypted conversations are spared
Interestingly, Google clarified that it’s only the unencrypted messages used to train AI on spam filtering. By deductive reasoning, this means all encrypted chats you have with friends and family, including RCS chats, are only subject to on-device machine learning assessments for potential spam. We also spotted the tech titan appending “and carries” to a line explaining who you can report spam to. This suggests you can now flag unwanted messages as spam to both Google and your cellular service provider, making it a much more effective way of filtering region-specific scams or attempts to target customers of a specific operator.
If the way Spam protection works has changed, it should be effective immediately because Google hasn’t shared any timelines or explicitly stated that anything is changing. In either case, we are grateful for the additional transparency about AI’s role in it all.