Dataverse - Anomalous application user activity

Status: available
Severity: medium
Time window: 14d
Group by: AnomalyTimeGenerated, InstanceUrl, OriginalObjectId, UserId
Source: github.com/Azure/Azure-Sentinel

Identifies anomalies in activity patterns of Dataverse application (non-interactive) users, based on activity falling outside the normal pattern of use.

MITRE ATT&CK coverage

Tactic	Techniques
Execution	`T1569` System Services
Credential Access	`T1528` Steal Application Access Token
Execution	`T0834` Native API `T0871` Execution through API
Persistence	`T0859` Valid Accounts
Lateral Movement	`T0859` Valid Accounts

Telemetry coverage

Platform	Record / event type
Microsoft 365	any: Dataverse and model-driven apps audit (catch-all)

Rule body

id: 0820da12-e895-417f-9175-7c256fcfb33e
kind: Scheduled
name: Dataverse - Anomalous application user activity
description: Identifies anomalies in activity patterns of Dataverse application (non-interactive)
  users, based on activity falling outside the normal pattern of use.
severity: Medium
status: Available
requiredDataConnectors:
  - connectorId: Dataverse
    dataTypes:
      - DataverseActivity
queryFrequency: 5h
queryPeriod: 14d
triggerOperator: gt
triggerThreshold: 0
tactics:
  - CredentialAccess
  - Execution
  - Persistence
relevantTechniques:
  - T1528
  - T1569
  - T0871
  - T0834
  - T0859
query: |
  let query_lookback = 14d;
  let query_frequency = 5h;
  let anomaly_threshold = 2.5;
  let seasonality = -1;
  let trend = 'linefit';
  let step_duration = 5h;
  let app_user_regex = "^[0-9A-Fa-f]{8}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{12}\\.com$";
  let guid_regex = "([0-9A-Fa-f]{8}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{12})";
  let application_users = DataverseActivity
      | where TimeGenerated >= ago(query_frequency)
      | where UserId !endswith "@onmicrosoft.com" and UserId != "Unknown"
      | summarize by UserId
      | where split(UserId, "@")[1] matches regex app_user_regex;
  DataverseActivity
  | where TimeGenerated >= startofday(ago(query_lookback))
  | where UserId in (application_users)
  | where isnotempty(OriginalObjectId)
  | make-series TotalEvents = count() default=0 on TimeGenerated from startofday(ago(query_lookback)) to now() step step_duration by UserId, InstanceUrl, OriginalObjectId
  | extend (Anomalies, Score, Baseline) = series_decompose_anomalies(TotalEvents, anomaly_threshold, seasonality, trend)
  | mv-expand
      TotalEvents to typeof(double),
      AnomalyTimeGenerated = TimeGenerated to typeof(datetime),
      Anomalies to typeof(double),
      Score to typeof(double),
      Baseline to typeof(long)
  | where Anomalies > 0
  | extend Details = bag_pack(
                         "TotalEvents",
                         TotalEvents,
                         "Anomalies",
                         Anomalies,
                         "Baseline",
                         Baseline,
                         "Score",
                         Score,
                         "OriginalObjectId",
                         OriginalObjectId
                     )
  | summarize Details = make_set(Details, 100) by UserId, InstanceUrl, AnomalyTimeGenerated
  | extend
      CloudAppId = int(32780),
      AadUserId = extract(guid_regex, 1, tostring(split(UserId, "@")[0]))
  | project
      AnomalyTimeGenerated,
      UserId,
      AadUserId,
      InstanceUrl,
      Details,
      CloudAppId
eventGroupingSettings:
  aggregationKind: AlertPerResult
entityMappings:
  - entityType: Account
    fieldMappings:
      - identifier: AadUserId
        columnName: AadUserId
  - entityType: CloudApplication
    fieldMappings:
      - identifier: AppId
        columnName: CloudAppId
      - identifier: InstanceName
        columnName: InstanceUrl
alertDetailsOverride:
  alertDisplayNameFormat: 'Dataverse - Non-interactive account anomaly detected in
    {{InstanceUrl}} '
  alertDescriptionFormat: 'Anomaly detected on {{UserId}} in {{InstanceUrl}}.  Details:
    {{Details}}'
customDetails:
  InstranceUrl: InstanceUrl
version: 3.2.0

Stages and Predicates

Parameters

let query_lookback = 14d;
let query_frequency = 5h;
let anomaly_threshold = 2.5;
let seasonality = -1;
let trend = 'linefit';
let step_duration = 5h;
let app_user_regex = "^[0-9A-Fa-f]{8}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{12}\\.com$";
let guid_regex = "([0-9A-Fa-f]{8}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{4}-[0-9A-Fa-f]{12})";

Let binding: `application_users` used in Stage 3

let application_users = DataverseActivity
    | where TimeGenerated >= ago(query_frequency)
    | where UserId !endswith "@onmicrosoft.com" and UserId != "Unknown"
    | summarize by UserId
    | where split(UserId, "@")[1] matches regex app_user_regex;

Stage 1: `source`

DataverseActivity

Stage 2: `where`

| where TimeGenerated >= startofday(ago(query_lookback))

Stage 3: `where`

| where UserId in (application_users)

Stage 4: `where`

| where isnotempty(OriginalObjectId)

The stages below score time-series anomalies (make-series, series_decompose_anomalies).

Stage 5: `make-series`

| make-series TotalEvents = count() default=0 on TimeGenerated from startofday(ago(query_lookback)) to now() step step_duration by UserId, InstanceUrl, OriginalObjectId

Stage 6: `extend`

| extend (Anomalies, Score, Baseline) = series_decompose_anomalies(TotalEvents, anomaly_threshold, seasonality, trend)

Stage 7: `mv-expand`

| mv-expand
    TotalEvents to typeof(double),
    AnomalyTimeGenerated = TimeGenerated to typeof(datetime),
    Anomalies to typeof(double),
    Score to typeof(double),
    Baseline to typeof(long)

Stage 8: `where`

| where Anomalies > 0

Stage 9: `extend`

| extend Details = bag_pack(
                       "TotalEvents",
                       TotalEvents,
                       "Anomalies",
                       Anomalies,
                       "Baseline",
                       Baseline,
                       "Score",
                       Score,
                       "OriginalObjectId",
                       OriginalObjectId
                   )

Stage 10: `summarize`

| summarize Details = make_set(Details, 100) by UserId, InstanceUrl, AnomalyTimeGenerated

Stage 11: `extend`

| extend
    CloudAppId = int(32780),
    AadUserId = extract(guid_regex, 1, tostring(split(UserId, "@")[0]))

Stage 12: `project`

| project
    AnomalyTimeGenerated,
    UserId,
    AadUserId,
    InstanceUrl,
    Details,
    CloudAppId

Indicators

These rows show field, operator, and value matches.

Field	Kind	Values	Search
`Anomalies`	gt	`0`	`field:"Anomalies" kind:gt value:"0"`
`OriginalObjectId`	is_not_null	(no value, null check)	`field:"OriginalObjectId" kind:is_not_null`
`UserId`	in	`application_users`	`field:"user" kind:in value:"application_users"`

Output fields

These fields are emitted when the rule matches.

Field	Source
`AadUserId`	`project`
`AnomalyTimeGenerated`	`project`
`CloudAppId`	`project`
`Details`	`project`
`InstanceUrl`	`project`
`UserId`	`project`

Dataverse - Anomalous application user activity

MITRE ATT&CK coverage

Telemetry coverage

Rule body

Stages and Predicates

Parameters

Let binding: application_users used in Stage 3

Stage 1: source

Stage 2: where

Stage 3: where

Stage 4: where

Stage 5: make-series

Stage 6: extend

Stage 7: mv-expand

Stage 8: where

Stage 9: extend

Stage 10: summarize

Stage 11: extend

Stage 12: project