Gmail Embraces Machine Learning

[Exterminator]

Google has rolled out new security features for Gmail customers, including early phishing detection using machine learning, click-time warnings for malicious links and unintended external reply warnings.

The new machine learning models in Gmail are based on a dedicated service that selectively delays messages (less than 0.05% of messages on average) to perform rigorous phishing analysis and further protect user data from compromise. This helps block spam and phishing messages from showing up in the inbox with over 99.9% accuracy, according to Andy Wen, senior product manager for Counter Abuse Technology at Gmail.

“This is huge, given that 50 to 70% of messages that Gmail receives are spam,” he said.

The detection models also integrate with Google Safe Browsing’s machine learning technologies for finding and flagging “phishy” and suspicious URLs, and then combine a variety of techniques (such as reputation and similarity analysis on URLs), allowing Gmail to generate new URL click-time warnings for phishing and malware links.

“As we find new patterns, our models adapt more quickly than manual systems ever could, and get better with time,” Wen said.

On the warnings front, Gmail for Work now displays unintended external reply warnings to users to help prevent data loss. For instance, if a user tries to respond to someone outside of her company domain, she will receive a quick warning to make sure she intended to send that email. It's a good first line of defense against imposter campaigns, like business email compromise/whaling attacks.

“Because Gmail has contextual intelligence, it knows if the recipient is an existing contact or someone you interact with regularly, to avoid displaying warnings unnecessarily,” Wen explained. “When employees are empowered to make the right decisions to protect data, it can improve an enterprise’s security posture.”

Gmail also has now built in defenses against ransomware and polymorphic malware, by combining thousands of spam, malware and ransomware signals with attachment heuristics (emails that could be threats based on signals) and sender signatures (already marked malware).

Other new features include the implementation of hosted S/MIME, to encrypt email while in transit; the Data Loss Prevention for Gmail service to protect sensitive information, and alerts when TLS encryption between mailboxes is not supported or when a message can’t be authenticated

 

[frogboy]

A very good implementation.

 

[XhenEd]

I bet Google has already been using machine learning for Gmail. Therefore, the title of the article is misleading if not wrong (the things they do for click-baiting).

What they do now is to implement a new strategy to combat phishing and other attacks using "new machine learning models".

Anyway, this is good news. It's always good to have new developments not only in combating cyber-security threats, but also in machine learning in general.

 

[onreact]

To be honest the biggest security issue with Gmail for advanced users is not phishing or malware. It's Google's own surveillance and advertising combination.

I read mail in plain text, don't click on links from people i don't know, don't open attachments in formats that could contain viruses (think .doc) etc. and most of the time I'm OK.

When using Gmail I do not open it in Gmail itself in order not to get bombarded with personalized ads that follow me around or feed the Google monster.

Ideally you use a secure and private mail provider like Proton Mail to break free from the Google tracking and monetization of your messages.

 

[Fritz]

So true. Like they haven't been analyzing every single word of every mail their users have ever sent or received for ages. LOL.

 

[Winter Soldier]

Good move but...Google is training its AI systems to read and understand the emails that we receive.
It would be helpful to know where the automatic analysis of addresses, names, locations in our emails, can push itself.
Information that, evidently, the artificial intelligence of Google is learning, but at the moment we do not know if this AI is used also for "extended" purposes over that fighting spam, fraud, etc.