What is UEBA and how does it spot odd sign-ins?

The short answer

UEBA (User and Entity Behavior Analytics) is software that learns what's normal for each person and device, then flags activity that doesn't fit, even when the password is correct.

Animated diagram: What is UEBA and how does it spot odd sign-ins?

UEBA

A stolen password is one of the quietest ways into an account. The attacker doesn’t need to break in. They simply sign in, and to the system the login looks like the real owner’s. Security people call this using valid accounts, and it’s hard to catch because the credentials themselves are correct.

UEBA (User and Entity Behavior Analytics) tackles that gap by looking at behavior instead of the secret. In Microsoft Sentinel, it keeps a behavior profile for users, hosts, IP (Internet Protocol) addresses, applications and other entities. When new activity arrives, it’s checked against those profiles, and anything that doesn’t fit stands out.

What UEBA actually compares

The raw material is activity logs, not message content. Sentinel draws on SigninLogs and AuditLogs, SecurityEvent records 4624, 4625, 4648, 4672 and 4688, AWS (Amazon Web Services) CloudTrail ConsoleLogin events, and Okta. VPN (virtual private network) and proxy events from Check Point, Fortinet and Zscaler come in through CommonSecurityLog.

From that history, each check keeps its own look-back window. A country you rarely connect from is judged over 10 days. A first-time internet provider, or ISP (internet service provider), over 30 days. A first-time country over 90 days. A first-time action, a first-time app and a dormant account over 180 days. The results land as flags such as FirstTimeUserConnectedFromCountry and FirstTimeUserConnectedViaISP, grouped inside ActivityInsights.

How UEBA works through one sign-in

  1. The event is logged. A sign-in, logon or cloud action is recorded with details such as the country, held in SourceIPLocation.
  2. It meets your baseline. The event is checked against days to months of your own history to see whether it’s new for you.
  3. Peers and company weigh in. UEBA also compares you with your top 20 peers, picked from security groups, mailing lists and other links using TF-IDF (term frequency-inverse document frequency) weighting, and with the whole organization.
  4. A priority score comes out. The InvestigationPriority field runs from 0, benign, to 10, highly anomalous. It mixes an Entity Anomaly Score for rarity with a Time Series Score for changes over time, like a spike in failed sign-ins.

The outcome is a number that tells an analyst where to look first. In the fictional Claire case, a first-time country and a new provider together push the score high, and Microsoft’s own example also gives first-time events a high score.

A high score is not a verdict

It’s tempting to read a high score as proof of a break-in. Microsoft only says higher scores are more likely to be true anomalies. Real life is messy: a business trip makes an honest sign-in look odd too, and Google lists travel as an innocent reason for an unfamiliar location. The score’s job is to get a person to check, not to decide on its own.

What to do

  • Type myaccount.google.com yourself, open Security, then Recent security events, and review what’s listed. On a phone the section is called Security & sign-in.
  • If an event wasn’t you, use Secure your account from that same review.
  • Under Your devices, open Manage all devices and look for any phone or computer you don’t recognize.
  • Remember the travel case: if you signed in from abroad, an alert about it may simply be your own trip.

#UEBA#BehaviorAnalytics#MicrosoftSentinel#IdentitySecurity#ThreatDetection#SIEM#AccountSecurity#CyberSecurity#InfoSec

Watch it in 60 seconds