Skip to main content

What is UEBA (user and entity behavior analytics)?

UEBA learns normal behavior for users, hosts, and apps, then flags activity that deviates from it, such as a compromised account or insider threat.

Glossary
Updated 2 min read

Definition

UEBA

Also called User and entity behavior analytics, UBA

UEBA (user and entity behavior analytics) is a detection technique that builds a baseline of normal activity for each user, host, application, and other entity, then flags behavior that deviates from that baseline as a possible threat.

How it works

UEBA models what "normal" looks like per entity: which systems a user signs in to, from where, at what times, how much data they move, and how they compare with peers in the same team. Microsoft Sentinel's UEBA uses machine learning to build behavioral profiles for users, hosts, IP addresses, applications, and other entities, then detects anomalies by comparing current activity with those baselines. It scores each anomaly using the entities involved, its severity, and context such as peer-group behavior and how the activity compares with the entity's own history.

The output is usually a risk score or an anomaly record rather than a definitive alert, which analysts and detection rules use to prioritize.

Common features and the questions behind them:

FeatureBaseline question
Sign-in location and deviceHas this user signed in from this country, network, or device before?
First-time accessHas this account ever touched this host, app, or cloud resource?
Data volumeHow much does this user normally download or upload per day?
Time of activityIs this outside the user's usual working hours?
Peer comparisonDo others in the same team or role do this?

A single deviation is weak evidence. Most teams get value by requiring several at once, for example a first-time host plus an unusual hour plus a new device, or by feeding anomalies into detections as context rather than alerting on each one.

Baseline choices are the main tuning knobs: how much history to learn from, which statistic defines normal, and how far above it counts as unusual. A median resists one-off spikes better than an average, and a longer window smooths over vacations and month-end peaks.

Example

A contractor who normally downloads a handful of files a week pulls hundreds from a shared drive in the days before their engagement ends. No rule names that file count, but it is far outside both their own history and their peers', so UEBA raises their risk score and an analyst checks whether the downloads were approved.

Why it matters for a SOC

Signature and indicator rules catch known-bad activity. UEBA covers attacks that use legitimate credentials and tools, such as compromised accounts, insider threats, and lateral movement, where nothing looks malicious except the pattern. The cost is noise: baselines drift, and unusual isn't the same as malicious, so raw anomalies produce false positives unless they're tuned and combined with context.

Baselines also need fast queries over long history. Cotool's SIEM, for instance, runs aggregations and baselines to power anomaly and behavioral detections and lets you write them as Python. Its detection page shows one for exfiltration to personal storage that models each user's normal upload volume as a 30-day rolling median and flags days over 20× that baseline.

Go deeper

FAQ

Is UEBA part of a SIEM?
Often. UEBA started as standalone products and is now usually a SIEM or XDR feature. Microsoft Sentinel, for example, builds UEBA behavioral profiles from the data connected to the workspace.

Sources

  1. Advanced threat detection with User and Entity Behavior Analytics (UEBA) in Microsoft Sentinel · accessed November 16, 2026
  2. AI-Native SIEM Built for Agents | Cotool · accessed November 16, 2026
  3. AI Detection Engineering with Python Detections | Cotool · accessed November 16, 2026

Attackers are scaling with tokens

Cotool helps defenders operate at machine speed. See how security teams are scaling Detection & Response.

Book a demo