Malformed user agent

Hunt Hypothesis

Malformed user agent strings may indicate the presence of malware that hardcodes these values to evade detection. SOC teams should proactively hunt for this behavior in Azure Sentinel to identify potential malware activity and mitigate associated risks.

KQL Query

(union isfuzzy=true
(OfficeActivity | where UserAgent != ""),
(OfficeActivity
| where RecordType in ("AzureActiveDirectory", "AzureActiveDirectoryStsLogon")
| extend OperationName = Operation
| parse ExtendedProperties with * 'User-Agent\\":\\"' UserAgent2 '\\' *
| parse ExtendedProperties with * 'UserAgent",      "Value": "' UserAgent1 '"' *
| where isnotempty(UserAgent1) or isnotempty(UserAgent2)
| extend UserAgent = iff( RecordType == 'AzureActiveDirectoryStsLogon', UserAgent1, UserAgent2)
| summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = ClientIP, Account = UserId, Type, RecordType, Operation
),
(AzureDiagnostics
| where ResourceType =~ "APPLICATIONGATEWAYS"
| where OperationName =~ "ApplicationGatewayAccess"
| extend ClientIP = columnifexists("clientIP_s", "None"), UserAgent = columnifexists("userAgent_s", "None")
| where UserAgent != '-'
| summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = ClientIP,  requestUri_s, httpMethod_s, host_s, requestQuery_s, Type
),
(
W3CIISLog
| where isnotempty(csUserAgent)
| summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent = csUserAgent, SourceIP = cIP, Account = csUserName, Type, sSiteName, csMethod, csUriStem
),
(
AWSCloudTrail
| where isnotempty(UserAgent)
| summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = SourceIpAddress, Account = UserIdentityUserName, Type, EventSource, EventName
),
(SigninLogs
| where isnotempty(UserAgent)
| summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = IPAddress, Account = UserPrincipalName, Type, OperationName, tostring(LocationDetails), tostring(DeviceDetail), AppDisplayName, ClientAppUsed
),
(AADNonInteractiveUserSignInLogs
| where isnotempty(UserAgent)
| summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = IPAddress, Account = UserPrincipalName, Type, OperationName, tostring(LocationDetails), tostring(DeviceDetail), AppDisplayName, ClientAppUsed
)
)
// Likely artefact of hardcoding
| where UserAgent startswith "User" or UserAgent startswith '\"'
// Incorrect casing
or (UserAgent startswith "Mozilla" and not(UserAgent contains_cs "Mozilla"))
// Incorrect casing
or UserAgent contains_cs  "(Compatible;"
// Missing MSIE version
or UserAgent matches regex @"MSIE\s?;"
// Incorrect spacing around MSIE version
or UserAgent matches regex  @"MSIE(?:\d|.{1,5}?\d\s;)"
| extend AccountName = split(Account, "@")[0], UPNSuffix = split(Account, "@")[1]

Analytic Rule Definition

id: a357535e-f722-4afe-b375-cff362b2b376
name: Malformed user agent
description: |
  'Malware authors will sometimes hardcode user agent string values when writing the network communication component of their malware. Malformed user agents can be an indication of such malware.'
severity: Medium
requiredDataConnectors:
  - connectorId: WAF
    dataTypes:
      - AzureDiagnostics
  - connectorId: Office365
    dataTypes:
      - OfficeActivity
  - connectorId: AzureActiveDirectory
    dataTypes:
      - SigninLogs
  - connectorId: AzureActiveDirectory
    dataTypes:
      - AADNonInteractiveUserSignInLogs
  - connectorId: AWS
    dataTypes:
      - AWSCloudTrail
  - connectorId: AzureMonitor(IIS)
    dataTypes:
      - W3CIISLog
queryFrequency: 1d
queryPeriod: 1d
triggerOperator: gt
triggerThreshold: 0
tactics:
  - InitialAccess
  - CommandAndControl
  - Execution
relevantTechniques:
  - T1189
  - T1071
  - T1203
query: |
  (union isfuzzy=true
  (OfficeActivity | where UserAgent != ""),
  (OfficeActivity
  | where RecordType in ("AzureActiveDirectory", "AzureActiveDirectoryStsLogon")
  | extend OperationName = Operation
  | parse ExtendedProperties with * 'User-Agent\\":\\"' UserAgent2 '\\' *
  | parse ExtendedProperties with * 'UserAgent",      "Value": "' UserAgent1 '"' *
  | where isnotempty(UserAgent1) or isnotempty(UserAgent2)
  | extend UserAgent = iff( RecordType == 'AzureActiveDirectoryStsLogon', UserAgent1, UserAgent2)
  | summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = ClientIP, Account = UserId, Type, RecordType, Operation
  ),
  (AzureDiagnostics
  | where ResourceType =~ "APPLICATIONGATEWAYS"
  | where OperationName =~ "ApplicationGatewayAccess"
  | extend ClientIP = columnifexists("clientIP_s", "None"), UserAgent = columnifexists("userAgent_s", "None")
  | where UserAgent != '-'
  | summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = ClientIP,  requestUri_s, httpMethod_s, host_s, requestQuery_s, Type
  ),
  (
  W3CIISLog
  | where isnotempty(csUserAgent)
  | summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent = csUserAgent, SourceIP = cIP, Account = csUserName, Type, sSiteName, csMethod, csUriStem
  ),
  (
  AWSCloudTrail
  | where isnotempty(UserAgent)
  | summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = SourceIpAddress, Account = UserIdentityUserName, Type, EventSource, EventName
  ),
  (SigninLogs
  | where isnotempty(UserAgent)
  | summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = IPAddress, Account = UserPrincipalName, Type, OperationName, tostring(LocationDetails), tostring(DeviceDetail), AppDisplayName, ClientAppUsed
  ),
  (AADNonInteractiveUserSignInLogs
  | where isnotempty(UserAgent)
  | summarize StartTime = min(TimeGenerated), EndTime = max(TimeGenerated) by UserAgent, SourceIP = IPAddress, Account = UserPrin

Required Data Sources

MITRE ATT&CK Context

• Tactic: InitialAccess
• Tactic: CommandAndControl
• Tactic: Execution
• Technique: T1189 — T1189
• Technique: T1071 — T1071
• Technique: T1203 — T1203

References

• Source Query

False Positive Guidance

• Scenario: A legitimate system update tool (e.g., Windows Update or Chocolatey) sends a malformed user agent string during a scheduled job.
  Filter/Exclusion: Exclude traffic from known update tools or IP ranges associated with system update services.

• Scenario: A network monitoring tool (e.g., Wireshark, tcpdump, or Nmap) generates malformed user agent strings when performing packet capture or port scanning activities.
  Filter/Exclusion: Exclude traffic originating from known network analysis tools or with specific command-line arguments associated with packet capture.

• Scenario: An administrator manually tests a web application using a tool like curl or Postman with a custom or malformed user agent string for diagnostic purposes.
  Filter/Exclusion: Exclude traffic from known admin workstations or user agents associated with testing tools.

• Scenario: A scheduled backup job (e.g., Veeam, Commvault, or rsync) uses a custom user agent string that is malformed due to a misconfiguration.
  Filter/Exclusion: Exclude traffic from backup servers or jobs with known backup-related user agent patterns.

• Scenario: A legitimate API testing tool (e.g., Insomnia, Swagger UI, or Postman) sends malformed user agents during automated test runs.
  Filter/Exclusion: Exclude traffic from known API testing tools or IP addresses associated with development environments.