Google is using machinery lesson to help Android users avoid unwanted and dangerous notifications

Notifications are part and packages of life, being useful and interesting. But they can also be problematic. Google is aware that notifications created by websites can be “spammy or even misleading”.

The worst offenders can try to deceive people in sharing personal information or downloading dangerous software. Google is using machinery lesson to fight again, starting warnings about adverse notifications warnings.

See also:

The characteristic is specific to the Android version of Chrome, aiming to separate internet notifications really useful from those that are meaningless or a security risk. The decision on how to categorize the notifications deals with the help of teaching machinery on the device.

In a blog post about the new feature, Google says:

When an announcement is flaged by Chrome, you will see the name of the site by sending the notice, a message warning that the content of the notice is potentially fraudulent or spammy, and the option to deregister from the site or to see the contents of the flag.

Google explains that machinery learning patterns are trained in the textual content of the notice, such as title texts, body and action button. The company points out that intimacy is integrated:

Notifications are completed to complete coded. The analysis of each message is done on the device and the notice content is not sent to Google to protect the user privacy. Due to the sensitive nature of the notifications content, the model was trained using synthetic data created by the large Gemini (LLM) model. The training data was evaluated against real announcements the Chrome security team collected by reconciling a variety of websites that were subsequently classified by human experts.

For now, the feature will be limited to the Android version of Chrome because Google says, “Most notifications are sent to mobile devices”. However, there is a chance that this same feature will eventually expand to other platforms, although there is no indication when.