Detection rules › Elastic
Azure Run Command Script Child Process
Identifies process start events whose parent matches Azure Virtual Machine Run Command execution patterns on Windows or Linux. On Windows, Run Command often launches PowerShell with -ExecutionPolicy Unrestricted and a script?.ps1 file; on Linux, the Azure Linux Agent (waagent) runs downloaded script.sh under "/var/lib/waagent/run-command/". Child process telemetry exposes the on-guest payload that cloud activity logs do not fully describe.
MITRE ATT&CK coverage
Rule body elastic
[metadata]
creation_date = "2026/05/20"
integration = ["endpoint", "windows"]
maturity = "production"
updated_date = "2026/05/20"
[rule]
author = ["Elastic"]
description = """
Identifies process start events whose parent matches Azure Virtual Machine Run Command execution patterns on Windows
or Linux. On Windows, Run Command often launches PowerShell with `-ExecutionPolicy Unrestricted` and a `script?.ps1`
file; on Linux, the Azure Linux Agent (waagent) runs downloaded script.sh under "/var/lib/waagent/run-command/".
Child process telemetry exposes the on-guest payload that cloud activity logs do not fully describe.
"""
false_positives = [
"""
Legitimate configuration management, extension deployment, or automation that uses Azure Run Command with the same
PowerShell or shell script paths may match. Baseline approved VM names, script naming, and deployment windows
before tuning.
""",
]
from = "now-9m"
index = [
"logs-endpoint.events.process*",
"logs-windows.sysmon_operational-*"
]
language = "eql"
license = "Elastic License v2"
name = "Azure Run Command Script Child Process"
note = """## Triage and analysis
### Investigating Azure Run Command Script Child Process
Azure VM Run Command executes scripts on guests without interactive RDP or SSH. On Windows, a parent PowerShell
process with `-ExecutionPolicy Unrestricted -File script?.ps1` often precedes child utilities; on Linux, `waagent`
invokes `/var/lib/waagent/run-command/download/*/script.sh` via `bash`, `sh`, or `dash`.
Correlate with `logs-azure.activitylogs-*` for `MICROSOFT.COMPUTE/VIRTUALMACHINES/RUNCOMMAND/ACTION` when available.
### Possible investigation steps
- Review `process.command_line`, `process.name`, and `process.parent.command_line` or `process.parent.args`.
- Confirm whether the host is an Azure VM and whether Run Command was expected for that asset.
- Pivot on `host.name` or `host.id` for other suspicious process or network activity in the same window.
### False positive analysis
- Extension handlers, guest configuration, and patch orchestration may use the same parent patterns.
- Exclude known automation hosts or script paths after validating with platform teams.
### Response and remediation
- If unauthorized, review Azure RBAC on the VM and subscription, revoke compromised credentials, and isolate the guest.
- Collect endpoint artifacts and Azure activity logs for incident reporting.
"""
references = [
"https://docs.microsoft.com/en-us/azure/virtual-machines/run-command",
"https://hackingthe.cloud/azure/run-command-abuse/"
]
risk_score = 47
rule_id = "cb9554e8-9f31-41a8-b4f5-d82144e6dc33"
severity = "medium"
tags = [
"Domain: Cloud",
"Domain: Endpoint",
"OS: Linux",
"OS: Windows",
"Use Case: Threat Detection",
"Tactic: Execution",
"Data Source: Elastic Defend",
"Data Source: Sysmon",
"Data Source: Azure",
"Resources: Investigation Guide",
]
timestamp_override = "event.ingested"
type = "eql"
query = '''
process where event.type in ("start", "process_started") and
(
(process.parent.name == "powershell.exe" and
process.parent.command_line like "powershell -ExecutionPolicy Unrestricted -File script?.ps1") or
(process.parent.name in ("bash", "dash", "sh", "tcsh", "csh", "zsh", "ksh", "fish", "mksh", "busybox") and
process.parent.args like "/var/lib/waagent/run-command/download/*/script.sh")
)
'''
[[rule.threat]]
framework = "MITRE ATT&CK"
[[rule.threat.technique]]
id = "T1059"
name = "Command and Scripting Interpreter"
reference = "https://attack.mitre.org/techniques/T1059/"
[[rule.threat.technique.subtechnique]]
id = "T1059.001"
name = "PowerShell"
reference = "https://attack.mitre.org/techniques/T1059/001/"
[[rule.threat.technique.subtechnique]]
id = "T1059.004"
name = "Unix Shell"
reference = "https://attack.mitre.org/techniques/T1059/004/"
[[rule.threat.technique]]
id = "T1651"
name = "Cloud Administration Command"
reference = "https://attack.mitre.org/techniques/T1651/"
[rule.threat.tactic]
id = "TA0002"
name = "Execution"
reference = "https://attack.mitre.org/tactics/TA0002/"
Stages and Predicates
Stage 1: process
process where event.type in ("start", "process_started") and
(
(process.parent.name == "powershell.exe" and
process.parent.command_line like "powershell -ExecutionPolicy Unrestricted -File script?.ps1") or
(process.parent.name in ("bash", "dash", "sh", "tcsh", "csh", "zsh", "ksh", "fish", "mksh", "busybox") and
process.parent.args like "/var/lib/waagent/run-command/download/*/script.sh")
)
Indicators
Each row is a field, operator, and value that the rule matches. The corpus column counts how many other rules in the catalog look for the same combination: high numbers point to widely-used, community-vetted indicators. Blank or 1 shows that the indicator is specific to this rule.
| Field | Kind | Values |
|---|---|---|
event.type | in |
|
process.parent.args | wildcard |
|
process.parent.command_line | wildcard |
|
process.parent.name | eq |
|
process.parent.name | in |
|