Endpoint Discovery Commands Followed by IMDS Requests

Detects a process that spawns multiple distinct host, user or network discovery utilities and within a short time window, accesses the cloud instance metadata service (IMDS), either directly or through a child process. Network requests to additional metadata endpoints, respective of the provider also meet this criteria.

Elastic rule (View on GitHub)

  1[metadata]
  2creation_date = "2026/10/23"
  3integration = ["endpoint"]
  4maturity = "production"
  5min_stack_comments = "MV_CONTAINS was added to ES|QL in 9.2.0 and is not available on 8.19; 9.4.0 is the lowest supported 9.x release."
  6min_stack_version = "9.4.0"
  7updated_date = "2026/11/18"
  8
  9[rule]
 10author = ["Elastic"]
 11description = """
 12Detects a process that spawns multiple distinct host, user or network discovery utilities and within a short time
 13window, accesses the cloud instance metadata service (IMDS), either directly or through a child
 14process. Network requests to additional metadata endpoints, respective of the provider also meet this criteria.
 15"""
 16false_positives = [
 17    """
 18    Cloud bootstrap tooling, guest agents, configuration management and monitoring agents may run discovery commands
 19    and query the instance metadata service from the same parent process. Validate the parent executable and command
 20    line and exclude approved automation.
 21    """,
 22]
 23from = "now-6m"
 24interval = "5m"
 25language = "esql"
 26license = "Elastic License v2"
 27name = "Endpoint Discovery Commands Followed by IMDS Requests"
 28note = """## Triage and analysis
 29
 30> **Disclaimer**:
 31> This investigation guide was created using generative AI technology and has been reviewed to improve its accuracy and relevance. While every effort has been made to ensure its quality, we recommend validating the content and adapting it to suit your specific environment and operational needs.
 32
 33
 34### Investigating Endpoint Discovery Commands Followed by IMDS Requests
 35
 36Each result groups the activity of a single parent process on a host: the discovery commands it spawned
 37(`Esql.recon_command_values`) and the connections to the instance metadata service made by the process itself or by
 38one of its children. The correlation key is stored in `Esql.correlation_id`.
 39
 40### Possible investigation steps
 41
 42- Review `Esql.process_parent_executable_values` and `Esql.process_parent_command_line_values` to identify the process
 43  that drove the activity and determine whether it is an interactive shell, a script interpreter, a web application, or
 44  approved automation.
 45- Inspect `Esql.process_command_line_values` for the exact discovery commands and any metadata request paths such as
 46  `/latest/meta-data/iam/security-credentials/` (AWS), `/metadata/identity/oauth2/token` (Azure), or
 47  `/computeMetadata/v1/instance/service-accounts/` (GCP).
 48- Identify the user in `Esql.user_name_values` and the session context, and check whether the activity followed a
 49  remote login, an exploit, or the execution of a newly dropped file.
 50- Correlate the host and time window with cloud control-plane logs to determine whether instance role credentials or
 51  managed identity tokens were subsequently used from this host or from an external IP address.
 52- Look for follow-on behavior such as cloud CLI or SDK usage, credential staging to disk, or outbound connections to
 53  unfamiliar destinations.
 54
 55### False positive analysis
 56
 57- Cloud-init, cloud guest agents, configuration management tools (Puppet, Chef, Salt, Ansible), MDM agents, and
 58  monitoring agents can legitimately run discovery utilities and query the metadata service. Confirm the parent
 59  executable path and exclude approved tooling by path or command line.
 60- In-house bootstrap or service startup scripts may collect host details and read instance metadata. Validate the script
 61  location, owner, and execution schedule.
 62
 63### Response and remediation
 64
 65- Isolate the affected instance or host and terminate the offending process tree.
 66- Revoke or rotate any instance role credentials, managed identity tokens, or secrets that may have been retrieved from
 67  the metadata service, and review the permissions attached to the instance role.
 68- Remove any dropped scripts, binaries, or persistence mechanisms associated with the activity.
 69- Enforce IMDSv2 or the strongest metadata protections offered by the cloud provider and restrict metadata access to
 70  the processes that require it.
 71"""
 72references = [
 73    "https://docs.litellm.ai/blog/security-incident-oct-21-2026",
 74]
 75risk_score = 47
 76rule_id = "ee9f9ede-3537-47aa-a94a-206690cd5037"
 77severity = "medium"
 78tags = [
 79    "Domain: Endpoint",
 80    "OS: Linux",
 81    "OS: macOS",
 82    "Use Case: Threat Detection",
 83    "Tactic: Discovery",
 84    "Tactic: Credential Access",
 85    "Threat: Supply Chain", 
 86    "Threat: IMDS Credential Theft",
 87    "Data Source: Elastic Defend",
 88    "Resources: Investigation Guide",
 89    "Rule Type: ES|QL",
 90    "Platform: Linux",
 91    "Platform: macOS",
 92]
 93timestamp_override = "event.ingested"
 94type = "esql"
 95query = '''
 96FROM logs-endpoint.events.process-*, logs-endpoint.events.network-*
 97| WHERE host.os.type IN ("linux", "macos")
 98| WHERE
 99    (
100      MV_CONTAINS(event.action, "exec")
101      AND
102      process.name.caseless IN (
103        // Linux
104        "ip", "ss", "hostnamectl", "uptime", "uname", "whoami", "id", "hostname",
105        "netstat", "arp", "ifconfig", "route", "w", "who",
106        "last", "getent", "dmidecode",
107        // macOS
108        "sw_vers", "system_profiler", "scutil", "networksetup", "ioreg",
109        "sysctl", "dscl", "dscacheutil", "csrutil", "spctl", "users", "groups"
110      )
111    )
112    OR destination.ip IN ("169.254.169.254", "169.254.170.2", "169.254.170.23", "100.100.100.200", "168.63.129.16")
113| WHERE process.parent.executable IS NULL OR NOT (
114    process.parent.executable LIKE "/Library/Application Support/JAMF/*" OR
115    process.parent.executable LIKE "/usr/local/jamf/*" OR
116    process.parent.executable LIKE "/opt/puppetlabs/*" OR
117    process.parent.executable LIKE "/opt/chef/*" OR
118    process.parent.executable LIKE "/usr/lib/google-cloud-sdk/*" OR
119    process.parent.executable LIKE "/Applications/Docker.app/Contents/MacOS/*"
120  )
121| EVAL
122    is_recon = process.name.caseless IN (
123      "ip", "ss", "hostnamectl", "uptime", "uname", "whoami", "id", "hostname",
124      "netstat", "arp", "ifconfig", "route", "w", "who",
125      "last", "getent", "dmidecode",
126      "sw_vers", "system_profiler", "scutil", "networksetup", "ioreg",
127      "sysctl", "dscl", "dscacheutil", "csrutil", "spctl", "users", "groups"
128    ) AND MV_CONTAINS(event.category, "process"),
129    is_metadata = destination.ip IN ("169.254.169.254", "169.254.170.2", "169.254.170.23", "100.100.100.200", "168.63.129.16")
130// recon events correlate on their parent, metadata events on both themselves and their parent
131| EVAL
132    Esql.correlation_id = CASE(
133      is_recon,
134      process.parent.entity_id,
135      MV_APPEND(process.entity_id, process.parent.entity_id)
136    ),
137    recon_command = CASE(is_recon, process.name.caseless, null),
138    metadata_access = CASE(is_metadata, 1, 0)
139| MV_EXPAND Esql.correlation_id
140| WHERE Esql.correlation_id IS NOT NULL
141| STATS
142    Esql.host_name_values = VALUES(host.name),
143    Esql.host_ip_values = VALUES(host.ip),
144    Esql.process_name_values = VALUES(process.name),
145    Esql.process_command_line_values = VALUES(process.command_line),
146    Esql.process_executable_values = VALUES(process.executable),
147    Esql.process_entity_id_values = VALUES(process.entity_id),
148    Esql.process_parent_entity_id_values = VALUES(process.parent.entity_id),
149    Esql.process_parent_name_values = VALUES(process.parent.name),
150    Esql.process_parent_executable_values = VALUES(process.parent.executable),
151    Esql.process_parent_command_line_values = VALUES(process.parent.command_line),
152    Esql.user_name_values = VALUES(user.name),
153    Esql.destination_ip_values = VALUES(destination.ip),
154    Esql.destination_port_values = VALUES(destination.port),
155    Esql.network_transport_values = VALUES(network.transport),
156    Esql.recon_command_values = VALUES(recon_command),
157    Esql.recon_command_count_distinct = COUNT_DISTINCT(recon_command),
158    Esql.metadata_access_count = SUM(metadata_access),
159    Esql.first_seen = MIN(@timestamp),
160    Esql.last_seen = MAX(@timestamp)
161  BY host.id, Esql.correlation_id
162| WHERE Esql.recon_command_count_distinct >= 3
163    AND Esql.metadata_access_count > 0
164    AND DATE_DIFF("seconds", Esql.first_seen, Esql.last_seen) <= 10
165| KEEP host.id, Esql.*
166'''
167
168[[rule.threat]]
169framework = "MITRE ATT&CK"
170
171[[rule.threat.technique]]
172id = "T1016"
173name = "System Network Configuration Discovery"
174reference = "https://attack.mitre.org/techniques/T1016/"
175
176[[rule.threat.technique]]
177id = "T1033"
178name = "System Owner/User Discovery"
179reference = "https://attack.mitre.org/techniques/T1033/"
180
181[[rule.threat.technique]]
182id = "T1082"
183name = "System Information Discovery"
184reference = "https://attack.mitre.org/techniques/T1082/"
185
186[rule.threat.tactic]
187id = "TA0007"
188name = "Discovery"
189reference = "https://attack.mitre.org/tactics/TA0007/"
190
191[[rule.threat]]
192framework = "MITRE ATT&CK"
193
194[[rule.threat.technique]]
195id = "T1552"
196name = "Unsecured Credentials"
197reference = "https://attack.mitre.org/techniques/T1552/"
198
199[[rule.threat.technique.subtechnique]]
200id = "T1552.005"
201name = "Cloud Instance Metadata API"
202reference = "https://attack.mitre.org/techniques/T1552/005/"
203
204[rule.threat.tactic]
205id = "TA0006"
206name = "Credential Access"
207reference = "https://attack.mitre.org/tactics/TA0006/"

Triage and analysis

Disclaimer: This investigation guide was created using generative AI technology and has been reviewed to improve its accuracy and relevance. While every effort has been made to ensure its quality, we recommend validating the content and adapting it to suit your specific environment and operational needs.

Investigating Endpoint Discovery Commands Followed by IMDS Requests

Each result groups the activity of a single parent process on a host: the discovery commands it spawned (Esql.recon_command_values) and the connections to the instance metadata service made by the process itself or by one of its children. The correlation key is stored in Esql.correlation_id.

Possible investigation steps

  • Review Esql.process_parent_executable_values and Esql.process_parent_command_line_values to identify the process that drove the activity and determine whether it is an interactive shell, a script interpreter, a web application, or approved automation.
  • Inspect Esql.process_command_line_values for the exact discovery commands and any metadata request paths such as /latest/meta-data/iam/security-credentials/ (AWS), /metadata/identity/oauth2/token (Azure), or /computeMetadata/v1/instance/service-accounts/ (GCP).
  • Identify the user in Esql.user_name_values and the session context, and check whether the activity followed a remote login, an exploit, or the execution of a newly dropped file.
  • Correlate the host and time window with cloud control-plane logs to determine whether instance role credentials or managed identity tokens were subsequently used from this host or from an external IP address.
  • Look for follow-on behavior such as cloud CLI or SDK usage, credential staging to disk, or outbound connections to unfamiliar destinations.

False positive analysis

  • Cloud-init, cloud guest agents, configuration management tools (Puppet, Chef, Salt, Ansible), MDM agents, and monitoring agents can legitimately run discovery utilities and query the metadata service. Confirm the parent executable path and exclude approved tooling by path or command line.
  • In-house bootstrap or service startup scripts may collect host details and read instance metadata. Validate the script location, owner, and execution schedule.

Response and remediation

  • Isolate the affected instance or host and terminate the offending process tree.
  • Revoke or rotate any instance role credentials, managed identity tokens, or secrets that may have been retrieved from the metadata service, and review the permissions attached to the instance role.
  • Remove any dropped scripts, binaries, or persistence mechanisms associated with the activity.
  • Enforce IMDSv2 or the strongest metadata protections offered by the cloud provider and restrict metadata access to the processes that require it.

References

Related rules

to-top