Detection rules › Kusto

Detect unauthorized data transfers using timeseries anomaly (ASIM Web Session)

Status
available
Severity
medium
Time window
14d
Group by
AnomalyHour, BaselineBytesSentinMBperHour, EventProduct, TotalBytesSentinMBperHour, anomalies, exists, score, sumexist
Source
github.com/Azure/Azure-Sentinel

'This query utilizes built-in KQL anomaly detection algorithms to identify anomalous data transfers to public networks. It detects significant deviations from a baseline pattern, allowing the detection of sudden increases in data transferred to unknown public networks, which may indicate data exfiltration attempts. Investigating such anomalies is crucial. The score indicates the degree to which the data transfer deviates from the baseline value. A higher score indicates a greater deviation. The query's output provides an aggregated summary view of the traffic observed in the flagged anomaly hour, including unique combinations of source IP addresses, destination IP addresses, and port bytes sent. It may be necessary to run queries for individual source IP addresses from the provided 'SourceIPlist' to identify any suspicious activity that warrants further investigation'

MITRE ATT&CK coverage

TacticTechniques
Exfiltration

Rule body

id: 5965d3e7-8ed0-477c-9b42-e75d9237fab0
name: Detect unauthorized data transfers using timeseries anomaly (ASIM Web Session)
description: |
  'This query utilizes built-in KQL anomaly detection algorithms to identify anomalous data transfers to public networks. It detects significant deviations from a baseline pattern, allowing the detection of sudden increases in data transferred to unknown public networks, which may indicate data exfiltration attempts. Investigating such anomalies is crucial.
  The score indicates the degree to which the data transfer deviates from the baseline value. A higher score indicates a greater deviation. The query's output provides an aggregated summary view of the traffic observed in the flagged anomaly hour, including unique combinations of source IP addresses, destination IP addresses, and port bytes sent. It may be necessary to run queries for individual source IP addresses from the provided 'SourceIPlist' to identify any suspicious activity that warrants further investigation'
severity: Medium
status: Available 
tags:
  - Schema: WebSession
    SchemaVersion: 0.2.6
requiredDataConnectors: []
queryFrequency: 1d
queryPeriod: 14d
triggerOperator: gt
triggerThreshold: 0
tactics:
  - Exfiltration
relevantTechniques:
  - T1030
query: |
  let startTime = 14d;
  let endTime = 1d;
  let timeframe = 1h;
  let scorethreshold = 5;
  let bytessentperhourthreshold = 10;
  // calculate avg. eps(events per second)
  let eps = materialize(_Im_WebSession(starttime=ago(1d))
      | project TimeGenerated
      | summarize AvgPerSec = count() / 3600 by bin(TimeGenerated, 1h)
      | summarize round(avg(AvgPerSec))
      );
  let summarizationexist  = (
      union isfuzzy=true 
          (
          WebSession_Summarized_SrcIP_CL
          | where EventTime_t > ago(1d) 
          | project v = int(2)
          ),
          (
          print int(1) 
          | project v = print_0
          )
      | summarize maxv = max(v)
      | extend sumexist = (maxv > 1)
      );
  let TimeSeriesData = union isfuzzy=true 
          (
          (datatable(exists: int, sumexist: bool)[1, false]
          | where toscalar(eps) > 1000
          | join (summarizationexist) on sumexist)
          | join (
              _Im_WebSession(starttime=ago(2d), endtime=now())
              | project DstIpAddr, SrcBytes, TimeGenerated, EventProduct
              | where isnotempty(DstIpAddr)
                  and not(ipv4_is_private(DstIpAddr))
                  and isnotempty(SrcBytes)
              | summarize SrcBytesSum=tolong(sum(SrcBytes)) by EventProduct, bin(TimeGenerated, 1h)
              | extend EventTime = TimeGenerated, exists=int(1)
              )
              on exists
          | project-away exists*, maxv, sum*
          ),
          (
          (datatable(exists: int, sumexist: bool)[1, false]
          | where toscalar(eps) between (501 .. 1000)
          | join (summarizationexist) on sumexist)
          | join (
              _Im_WebSession(starttime=ago(3d), endtime=now())
              | project DstIpAddr, SrcBytes, TimeGenerated, EventProduct
              | where isnotempty(DstIpAddr)
                  and not(ipv4_is_private(DstIpAddr))
                  and isnotempty(SrcBytes)
              | summarize SrcBytesSum=tolong(sum(SrcBytes)) by EventProduct, bin(TimeGenerated, 1h)
              | extend EventTime = TimeGenerated, exists=int(1)
              )
              on exists
          | project-away exists*, maxv, sum*
          ),
          (
          (datatable(exists: int, sumexist: bool)[1, false]
          | where toscalar(eps) <= 500
          | join (summarizationexist) on sumexist)
          | join (
              _Im_WebSession(starttime=ago(4d), endtime=now())
              | project DstIpAddr, SrcBytes, TimeGenerated, EventProduct
              | where isnotempty(DstIpAddr)
                  and not(ipv4_is_private(DstIpAddr))
                  and isnotempty(SrcBytes)
              | summarize SrcBytesSum=tolong(sum(SrcBytes)) by EventProduct, bin(TimeGenerated, 1h)
              | extend EventTime = TimeGenerated, exists=int(1)
              )
              on exists
          | project-away exists*, maxv, sum*
          ),
          (
          WebSession_Summarized_SrcIP_CL
          | where EventTime_t between (ago(startTime) .. now())
          | where isnotempty(SrcBytes_d) and not(DstIPIsPrivate_b)
          | project
              SrcBytesSum=tolong(SrcBytes_d),
              EventTime=EventTime_t,
              EventProduct = EventProduct_s
          )
      | make-series TotalBytesSent = sum(SrcBytesSum) on EventTime from startofday(ago(startTime)) to startofday(now()) step timeframe by EventProduct;
  // TimeSeriesData block ends here
  //Take only anomalies in TimeSeriesData
  let TimeSeriesAnomalies = materialize(TimeSeriesData
      | extend (anomalies, score, baseline) = series_decompose_anomalies(TotalBytesSent, scorethreshold, -1, 'linefit')
      | mv-expand
          TotalBytesSent to typeof(long),
          EventTime to typeof(datetime),
          anomalies to typeof(double),
          score to typeof(double),
          baseline to typeof(long)
      | where anomalies > 0 and baseline > 0
      | extend AnomalyHour = EventTime
      | extend
          TotalBytesSentinMBperHour = round(((TotalBytesSent / 1024) / 1024), 2),
          BaselineBytesSentinMBperHour = round(((baseline / 1024) / 1024), 2),
          score = round(score, 2)
      | project
          EventProduct,
          AnomalyHour,
          TotalBytesSentinMBperHour,
          BaselineBytesSentinMBperHour,
          anomalies,
          score
      | where AnomalyHour between (startofday(ago(endTime)) .. startofday(now())) // Get TimeSeriesAnomalies in previous day
          );
  // TimeSeriesAlerts block end here
  let AnomalyHours = materialize (TimeSeriesAnomalies
      | project AnomalyHour);
  //Previous day aggregated per hour
  let PreviousDayLogs = 
      _Im_WebSession(starttime=startofday(ago(endTime)), endtime=startofday(now()))
      | where isnotempty(DstIpAddr) and isnotempty(SrcIpAddr) and isnotempty(SrcBytes)
      | where not(ipv4_is_private(DstIpAddr))
      | project
          TimeGenerated,
          DstIpAddr,
          SrcIpAddr,
          SrcBytes,
          DstBytes,
          DstPortNumber,
          EventProduct
      | extend DateHour = bin(TimeGenerated, timeframe) // create a new column and round to hour
      | where DateHour in (AnomalyHours) // Filter dataset to include only anomaly AnomalyHours
      | extend
          SentBytesinMB = ((SrcBytes / 1024) / 1024),
          ReceivedBytesinMB = ((DstBytes / 1024) / 1024)
      | summarize
          HourlyCount = count(),
          TimeGeneratedMax = arg_max(TimeGenerated, *),
          DestinationIPList = make_set(DstIpAddr, 100),
          DestinationPortList = make_set(DstPortNumber, 100),
          SentBytesinMB = tolong(sum(SentBytesinMB)),
          ReceivedBytesinMB = tolong(sum(ReceivedBytesinMB))
          by SrcIpAddr, EventProduct, TimeGeneratedHour = bin(TimeGenerated, timeframe)
      | where SentBytesinMB > bytessentperhourthreshold
      | sort by TimeGeneratedHour asc, SentBytesinMB desc
      | extend Rank=row_number(1, prev(TimeGeneratedHour) != TimeGeneratedHour) // Ranking the dataset per Hourly Partition
      | where Rank <= 10  // Selecting Top 10 records with Highest BytesSent in each Hour
      | project
          EventProduct,
          TimeGeneratedHour,
          TimeGeneratedMax,
          SrcIpAddr,
          DestinationIPList,
          DestinationPortList,
          SentBytesinMB,
          ReceivedBytesinMB,
          Rank,
          HourlyCount;
  // PreviousDayLogs block ends here
  TimeSeriesAnomalies
  | join kind = inner (PreviousDayLogs
      | extend AnomalyHour = TimeGeneratedHour)
      on EventProduct, AnomalyHour
  | sort by score desc
  | project
      EventProduct,
      AnomalyHour,
      TimeGeneratedMax,
      SrcIpAddr,
      DestinationIPList,
      DestinationPortList,
      SentBytesinMB,
      ReceivedBytesinMB,
      TotalBytesSentinMBperHour,
      BaselineBytesSentinMBperHour,
      score,
      anomalies,
      HourlyCount
  | summarize
      EventCount = sum(HourlyCount),
      startTimeUtc = min(TimeGeneratedMax),
      EndTimeUtc = max(TimeGeneratedMax),
      SentBytesinMB = sum(SentBytesinMB),
      ReceivedBytesinMB = sum(ReceivedBytesinMB),
      SourceIP = take_any(SrcIpAddr),
      SourceIPList = make_set(SrcIpAddr, 10),
      DestinationIPList = make_set(DestinationIPList, 100),
      DestinationPortList = make_set(DestinationPortList, 100)
      by
      AnomalyHour,
      TotalBytesSentinMBperHour,
      BaselineBytesSentinMBperHour,
      score,
      anomalies,
      EventProduct
  | project
      EventProduct,
      AnomalyHour,
      startTimeUtc,
      EndTimeUtc,
      SourceIP,
      SourceIPList,
      DestinationIPList,
      DestinationPortList,
      SentBytesinMB,
      ReceivedBytesinMB,
      TotalBytesSentinMBperHour,
      BaselineBytesSentinMBperHour,
      anomalies,
      score,
      EventCount
entityMappings:
  - entityType: IP
    fieldMappings:
      - identifier: Address
        columnName: SourceIP
eventGroupingSettings:
  aggregationKind: AlertPerResult
customDetails:
  EventCount: EventCount
  SourceIPList: SourceIPList
  DestinationIPList: DestinationIPList
  DestinationPortList: DestinationPortList
  SentBytesinMB: SentBytesinMB
  ReceivedBytesinMB: ReceivedBytesinMB
  anomalies: anomalies
  score: score
alertDetailsOverride:
  alertDisplayNameFormat: "IP address '{{SourceIP}}' is engaged in data transfers to a public network that exceeds usual levels"
  alertDescriptionFormat: "Please conduct a thorough investigation of each IPAddresses listed in SourceIPList: '{{SourceIPList}}' to identify any suspicious activities that may require further investigation. 'SourceIPList' include the top 10 client IP addresses that transmitted the highest amount of data during the anomalous hour"
version: 1.0.1
kind: Scheduled

Stages and Predicates

Stage 0: let

let startTime = 14d;
let endTime = 1d;
let timeframe = 1h;
let scorethreshold = 5;
let bytessentperhourthreshold = 10;
let eps = materialize(_Im_WebSession(starttime=ago(1d))
    | project TimeGenerated
    | summarize AvgPerSec = count() / 3600 by bin(TimeGenerated, 1h)
    | summarize round(avg(AvgPerSec))
    );
let summarizationexist  = (
    union isfuzzy=true 
        (
        WebSession_Summarized_SrcIP_CL
        | where EventTime_t > ago(1d) 
        | project v = int(2)
        ),
        (
        print int(1) 
        | project v = print_0
        )
    | summarize maxv = max(v)
    | extend sumexist = (maxv > 1)
    );
let TimeSeriesData = union <inlined as stages below>;
let TimeSeriesAnomalies = materialize(<inlined as stages below>);
let AnomalyHours = materialize (TimeSeriesAnomalies
    | project AnomalyHour);
let PreviousDayLogs = 
    _Im_WebSession(starttime=startofday(ago(endTime)), endtime=startofday(now()))
    | where isnotempty(DstIpAddr) and isnotempty(SrcIpAddr) and isnotempty(SrcBytes)
    | where not(ipv4_is_private(DstIpAddr))
    | project
        TimeGenerated,
        DstIpAddr,
        SrcIpAddr,
        SrcBytes,
        DstBytes,
        DstPortNumber,
        EventProduct
    | extend DateHour = bin(TimeGenerated, timeframe)
    | where DateHour in (AnomalyHours)
    | extend
        SentBytesinMB = ((SrcBytes / 1024) / 1024),
        ReceivedBytesinMB = ((DstBytes / 1024) / 1024)
    | summarize
        HourlyCount = count(),
        TimeGeneratedMax = arg_max(TimeGenerated, *),
        DestinationIPList = make_set(DstIpAddr, 100),
        DestinationPortList = make_set(DstPortNumber, 100),
        SentBytesinMB = tolong(sum(SentBytesinMB)),
        ReceivedBytesinMB = tolong(sum(ReceivedBytesinMB))
        by SrcIpAddr, EventProduct, TimeGeneratedHour = bin(TimeGenerated, timeframe)
    | where SentBytesinMB > bytessentperhourthreshold
    | sort by TimeGeneratedHour asc, SentBytesinMB desc
    | extend Rank=row_number(1, prev(TimeGeneratedHour) != TimeGeneratedHour)
    | where Rank <= 10
    | project
        EventProduct,
        TimeGeneratedHour,
        TimeGeneratedMax,
        SrcIpAddr,
        DestinationIPList,
        DestinationPortList,
        SentBytesinMB,
        ReceivedBytesinMB,
        Rank,
        HourlyCount;

Stage 1: union

union of 4 branches

Stage 2: source

datatable

Stage 3: where

where /* macro: (toscalar(eps) > 1000) */

Stage 4: join

join (summarizationexist) on sumexist

Stage 5: join

| join (
            _Im_WebSession(starttime=ago(2d), endtime=now())
            | project DstIpAddr, SrcBytes, TimeGenerated, EventProduct
            | where isnotempty(DstIpAddr)
                and not(ipv4_is_private(DstIpAddr))
                and isnotempty(SrcBytes)
            | summarize SrcBytesSum=tolong(sum(SrcBytes)) by EventProduct, bin(TimeGenerated, 1h)
            | extend EventTime = TimeGenerated, exists=int(1)
            )
            on exists

Stage 6: project-away

project-away exists*, maxv, sum*

Stage 7: source

datatable

Stage 8: where

where /* macro: (toscalar(eps) between (501 .. 1000)) */

Stage 9: join

join (summarizationexist) on sumexist

Stage 10: join

| join (
            _Im_WebSession(starttime=ago(3d), endtime=now())
            | project DstIpAddr, SrcBytes, TimeGenerated, EventProduct
            | where isnotempty(DstIpAddr)
                and not(ipv4_is_private(DstIpAddr))
                and isnotempty(SrcBytes)
            | summarize SrcBytesSum=tolong(sum(SrcBytes)) by EventProduct, bin(TimeGenerated, 1h)
            | extend EventTime = TimeGenerated, exists=int(1)
            )
            on exists

Stage 11: project-away

project-away exists*, maxv, sum*

Stage 12: source

datatable

Stage 13: where

where /* macro: (toscalar(eps) <= 500) */

Stage 14: join

join (summarizationexist) on sumexist

Stage 15: join

| join (
            _Im_WebSession(starttime=ago(4d), endtime=now())
            | project DstIpAddr, SrcBytes, TimeGenerated, EventProduct
            | where isnotempty(DstIpAddr)
                and not(ipv4_is_private(DstIpAddr))
                and isnotempty(SrcBytes)
            | summarize SrcBytesSum=tolong(sum(SrcBytes)) by EventProduct, bin(TimeGenerated, 1h)
            | extend EventTime = TimeGenerated, exists=int(1)
            )
            on exists

Stage 16: project-away

project-away exists*, maxv, sum*

Stage 17: source

WebSession_Summarized_SrcIP_CL

Stage 18: where

| where EventTime_t between (ago(startTime) .. now())

Stage 19: where

| where isnotempty(SrcBytes_d) and not(DstIPIsPrivate_b)

Stage 20: project

| project
            SrcBytesSum=tolong(SrcBytes_d),
            EventTime=EventTime_t,
            EventProduct = EventProduct_s

Stage 21: summarize

summarize TotalBytesSent by EventProduct

Stage 22: extend

extend anomalies, baseline, score

Stage 23: mv-expand

mv-expand TotalBytesSent

Stage 24: where

where anomalies > 0 and baseline > 0

Stage 25: extend

extend AnomalyHour

Stage 26: extend

extend BaselineBytesSentinMBperHour, TotalBytesSentinMBperHour, score

Stage 27: project

project AnomalyHour, BaselineBytesSentinMBperHour, EventProduct, TotalBytesSentinMBperHour, anomalies, score

Stage 28: where

where /* macro: (AnomalyHour between (startofday(ago(1d)) .. startofday(now()))) */

Stage 29: join

| join kind = inner (PreviousDayLogs
    | extend AnomalyHour = TimeGeneratedHour)
    on EventProduct, AnomalyHour

Stage 30: sort

sort by score

Stage 31: project

project AnomalyHour, BaselineBytesSentinMBperHour, DestinationIPList, DestinationPortList, EventProduct, HourlyCount, ReceivedBytesinMB, SentBytesinMB, SrcIpAddr, TimeGeneratedMax, TotalBytesSentinMBperHour, anomalies, score

Stage 32: summarize

summarize DestinationIPList, DestinationPortList, EndTimeUtc, EventCount, ReceivedBytesinMB, SentBytesinMB, SourceIP, SourceIPList, startTimeUtc by AnomalyHour, TotalBytesSentinMBperHour, BaselineBytesSentinMBperHour, score, anomalies, EventProduct

Stage 33: project

project AnomalyHour, BaselineBytesSentinMBperHour, DestinationIPList, DestinationPortList, EndTimeUtc, EventCount, EventProduct, ReceivedBytesinMB, SentBytesinMB, SourceIP, SourceIPList, TotalBytesSentinMBperHour, anomalies, score, startTimeUtc

Indicators

These rows show field, operator, and value matches.

Exclusions

The rule actively suppresses these predicates.

FieldKindExcluded valuesSearch
DstIpAddrcidr_match10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16, 169.254.0.0/16, 127.0.0.0/8excludes:DstIpAddr

Output fields

These fields are emitted when the rule matches.

FieldSource
AnomalyHourproject
BaselineBytesSentinMBperHourproject
DestinationIPListproject
DestinationPortListproject
EndTimeUtcproject
EventCountproject
EventProductproject
ReceivedBytesinMBproject
SentBytesinMBproject
SourceIPproject
SourceIPListproject
TotalBytesSentinMBperHourproject
anomaliesproject
scoreproject
startTimeUtcproject