Detection rules › Splunk
Kubernetes Previously Unseen Container Image Name
The following analytic identifies the creation of containerized workloads using previously unseen images in a Kubernetes cluster. It leverages process metrics from an OTEL collector and Kubernetes cluster receiver, pulled from Splunk Observability Cloud. The detection compares container image names seen in the last hour with those from the previous 30 days. This activity is significant as unfamiliar container images may introduce vulnerabilities, malware, or misconfigurations, posing threats to the cluster's integrity. If confirmed malicious, compromised images can lead to data breaches, service disruptions, unauthorized access, and potential lateral movement within the cluster.
MITRE ATT&CK coverage
| Tactic | Techniques |
|---|---|
| Execution | T1204 User Execution |
Rule body splunk
name: Kubernetes Previously Unseen Container Image Name
id: fea515a4-b1d8-4cd6-80d6-e0d71397b891
version: 10
creation_date: '2024-01-10'
modification_date: '2026-05-13'
author: Matthew Moore, Splunk
status: experimental
type: Anomaly
description: The following analytic identifies the creation of containerized workloads using previously unseen images in a Kubernetes cluster. It leverages process metrics from an OTEL collector and Kubernetes cluster receiver, pulled from Splunk Observability Cloud. The detection compares container image names seen in the last hour with those from the previous 30 days. This activity is significant as unfamiliar container images may introduce vulnerabilities, malware, or misconfigurations, posing threats to the cluster's integrity. If confirmed malicious, compromised images can lead to data breaches, service disruptions, unauthorized access, and potential lateral movement within the cluster.
data_source: []
search: |-
| mstats count(k8s.container.ready) as k8s.container.ready_count where `kubernetes_metrics` AND earliest=-24h by host.name k8s.cluster.name k8s.node.name container.image.name
| eval current="True"
| append [mstats count(k8s.container.ready) as k8s.container.ready_count where `kubernetes_metrics` AND earliest=-30d latest=-1h by host.name k8s.cluster.name k8s.node.name container.image.name
| eval current="false" ]
| stats values(current) as current
BY host.name k8s.cluster.name k8s.node.name
container.image.name
| search current="true" AND current!="false"
| rename host.name as host
| `kubernetes_previously_unseen_container_image_name_filter`
how_to_implement: "To implement this detection, follow these steps:\n* Deploy the OpenTelemetry Collector (OTEL) to your Kubernetes cluster.\n* Enable the hostmetrics/process receiver in the OTEL configuration.\n* Ensure that the process metrics, specifically Process.cpu.utilization and process.memory.utilization, are enabled.\n* Install the Splunk Infrastructure Monitoring (SIM) add-on. (ref: https://splunkbase.splunk.com/app/5247)\n * Configure the SIM add-on with your Observability Cloud Organization ID and Access Token.\n* Set up the SIM modular input to ingest Process Metrics. Name this input \"sim_process_metrics_to_metrics_index\".\n* In the SIM configuration, set the Organization ID to your Observability Cloud Organization ID.\n* Set the Signal Flow Program to the following: data('process.threads').publish(label='A'); data('process.cpu.utilization').publish(label='B'); data('process.cpu.time').publish(label='C'); data('process.disk.io').publish(label='D'); data('process.memory.usage').publish(label='E'); data('process.memory.virtual').publish(label='F'); data('process.memory.utilization').publish(label='G'); data('process.cpu.utilization').publish(label='H'); data('process.disk.operations').publish(label='I'); data('process.handles').publish(label='J'); data('process.threads').publish(label='K')\n* Set the Metric Resolution to 10000.\n * Leave all other settings at their default values.\n* Run the Search Baseline Of Kubernetes Container Network IO Ratio"
known_false_positives: No false positives have been identified at this time.
references:
- https://github.com/signalfx/splunk-otel-collector-chart
intermediate_findings:
entities:
- field: host
type: system
score: 20
message: Kubernetes Previously Unseen Container Image Name on host $host$
analytic_story:
- Abnormal Kubernetes Behavior using Splunk Infrastructure Monitoring
asset_type: Kubernetes
mitre_attack_id:
- T1204
product:
- Splunk Enterprise
- Splunk Enterprise Security
- Splunk Cloud
category: cloud
security_domain: network
Stages and Predicates
Stage 1: search
| mstats count(k8s.container.ready) as k8s.container.ready_count where `kubernetes_metrics` AND earliest=-24h by host.name k8s.cluster.name k8s.node.name container.image.name
Stage 2: eval
| eval current="True"
Stage 3: append
| append [mstats count(k8s.container.ready) as k8s.container.ready_count where `kubernetes_metrics` AND earliest=-30d latest=-1h by host.name k8s.cluster.name k8s.node.name container.image.name
| eval current="false" ]
Stage 4: stats
| stats values(current) as current
BY host.name k8s.cluster.name k8s.node.name
container.image.name
Stage 5: search
| search current="true" AND current!="false"
Stage 6: rename
| rename host.name as host
Stage 7: search
| `kubernetes_previously_unseen_container_image_name_filter`
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.
Search terms
Bare-string tokens in the SPL search body. Splunk matches each token against _raw (the untyped raw event text) anywhere it appears, not against a specific field. These don't surface in the Indicators table because they aren't predicates on a known field.
| Stage | Term |
|---|---|
| 1 | mstats |
| 1 | count |
| 1 | k8s.container.ready |
| 1 | as |
| 1 | k8s.container.ready_count |
| 1 | where |
| 1 | by |
| 1 | host.name |
| 1 | k8s.cluster.name |
| 1 | k8s.node.name |
| 1 | container.image.name |