GKE Sensitive RBAC Change Followed by Workload Modification

ELASTIC detection rule by Elastic · Elastic License v2

Detects when the same GKE identity creates or modifies a Role or ClusterRole with high-risk permissions (wildcard access, RBAC escalation verbs, or access to secrets / privileged APIs) and also creates or patches a DaemonSet, Deployment, or CronJob within five minutes. This correlation is consistent with RBAC-based privilege escalation followed by payload deployment.

[metadata]
creation_date = "2026/07/17"
integration = ["gcp"]
maturity = "production"
updated_date = "2026/08/25"

[rule]
author = ["Elastic"]
description = """
Detects when the same GKE identity creates or modifies a Role or ClusterRole with high-risk permissions
(wildcard access, RBAC escalation verbs, or access to secrets / privileged APIs) and also creates or patches
a DaemonSet, Deployment, or CronJob within five minutes. This correlation is consistent with RBAC-based
privilege escalation followed by payload deployment.
"""
false_positives = [
    """
    GitOps or platform controllers that sync sensitive Roles and immediately reconcile DaemonSets, Deployments,
    or CronJobs can match. Baseline approved automation after confirming the Role content is intentional.
    """,
    """
    Coordinated maintenance that expands RBAC and rolls workloads in the same change window may alert; validate
    against change tickets and the actor's expected client fingerprint.
    """,
]
from = "now-11m"
interval = "5m"
language = "esql"
license = "Elastic License v2"
name = "GKE Sensitive RBAC Change Followed by Workload Modification"
note = """## Triage and analysis

### Investigating GKE Sensitive RBAC Change Followed by Workload Modification

This ES|QL rule correlates two successful GKE audit behaviors from the same `client.user.email` within five
minutes:

1. Role or ClusterRole create/update/patch that grants high-risk permissions (wildcards, `escalate` /
   `bind` / `impersonate`, secret read, or privileged API resources such as `pods/exec` and
   `serviceaccounts/token`)
2. DaemonSet, Deployment, or CronJob create or patch after the sensitive RBAC change

`Esql.rbac_to_workload_minutes` is the gap from the latest sensitive RBAC event to the earliest workload
modification in the lookback window.

### Possible investigation steps

- Review `Esql.event_action_values` and `Esql.gcp_audit_resource_name_values` for the Role/ClusterRole and
  workload objects touched.
- Inspect `Esql.user_agent_original_values` and `Esql.source_ip_values` for unexpected clients or networks.
- Check for RoleBinding or ClusterRoleBinding activity around the same identity and time window.
- Correlate with secret access, pod exec, or token creation from the same actor.

### False positive analysis

- GitOps pipelines that manage both RBAC manifests and workloads in one sync cycle.
- Platform bootstrap that patches built-in roles and reconciles addon workloads.

### Response and remediation

- Roll back unauthorized Role/ClusterRole and workload changes, revoke the actor's credentials, and
  tighten who can mutate RBAC and sensitive workloads.
"""
setup = """
The GCP Fleet integration with GKE audit logs enabled is required. Request body capture on RBAC
resources is required so Role and ClusterRole rule verbs and resources are present in
`gcp.audit.request`.
"""
references = [
    "https://heilancoos.github.io/research/2025/12/16/kubernetes.html#overly-permissive-role-based-access-control",
    "https://kubernetes.io/docs/reference/access-authn-authz/rbac/",
]
risk_score = 47
rule_id = "3ec2175d-8f34-4be9-b3be-c5821538760e"
severity = "medium"
tags = [
    "Domain: Cloud",
    "Domain: Kubernetes",
    "Domain: Containers",
    "Data Source: GCP",
    "Data Source: Google Cloud Platform",
    "Data Source: GCP Audit Logs",
    "Platform: GCP",
    "Platform: Kubernetes",
    "Use Case: Threat Detection",
    "Tactic: Privilege Escalation",
    "Tactic: Persistence",
    "Resources: Investigation Guide",
    "Rule Type: ESQL",
]
timestamp_override = "event.ingested"
type = "esql"

query = '''
from logs-gcp.audit-* metadata _id, _index, _version
| where data_stream.dataset == "gcp.audit"
    and service.name == "k8s.io"
    and event.outcome == "success"
    and client.user.email is not null
    and source.ip is not null
    and not to_string(source.ip) in ("127.0.0.1", "::1")
    and not client.user.email in (
      "system:addon-manager",
      "system:apiserver",
      "system:kube-controller-manager",
      "system:serviceaccount:flux-system:kustomize-controller",
      "system:serviceaccount:flux-system:helm-controller",
      "system:serviceaccount:flux-system:source-controller",
      "system:serviceaccount:flux-system:flux-operator",
      "system:serviceaccount:flux-operator:flux-operator"
    )
    and (
      (
        event.action in (
          "io.k8s.authorization.rbac.v1.roles.create",
          "io.k8s.authorization.rbac.v1.roles.update",
          "io.k8s.authorization.rbac.v1.roles.patch",
          "io.k8s.authorization.rbac.v1.clusterroles.create",
          "io.k8s.authorization.rbac.v1.clusterroles.update",
          "io.k8s.authorization.rbac.v1.clusterroles.patch"
        )
        and not (
          client.user.email == "system:serviceaccount:kube-system:clusterrole-aggregation-controller"
          and event.action == "io.k8s.authorization.rbac.v1.clusterroles.patch"
        )
        and (
          KQL("""gcp.audit.request.rules.verbs:("*" or "escalate" or "bind" or "impersonate")""")
          or KQL("""gcp.audit.request.rules.verbs:("*" or "create" or "patch" or "update") and gcp.audit.request.rules.resources:("*" or "clusterroles" or "clusterrolebindings" or "roles" or "rolebindings" or "pods/exec" or "serviceaccounts/token" or "nodes/proxy" or "daemonsets")""")
          or KQL("""gcp.audit.request.rules.verbs:("*" or "get" or "list") and gcp.audit.request.rules.resources:("*" or "secrets")""")
        )
      )
      or event.action in (
        "io.k8s.apps.v1.daemonsets.create",
        "io.k8s.apps.v1.daemonsets.patch",
        "io.k8s.apps.v1.deployments.create",
        "io.k8s.apps.v1.deployments.patch",
        "io.k8s.batch.v1.cronjobs.create",
        "io.k8s.batch.v1.cronjobs.patch"
      )
    )
| eval Esql.is_sensitive_rbac = case(event.action like "io.k8s.authorization.rbac.*", 1, 0),
    Esql.is_workload_modification = case(event.action like "io.k8s.apps.*" or event.action like "io.k8s.batch.*", 1, 0),
    Esql.sensitive_rbac_timestamp = case(event.action like "io.k8s.authorization.rbac.*", @timestamp, null),
    Esql.workload_modification_timestamp = case(event.action like "io.k8s.apps.*" or event.action like "io.k8s.batch.*", @timestamp, null)
| stats
    Esql.sensitive_rbac_count = sum(Esql.is_sensitive_rbac),
    Esql.workload_modification_count = sum(Esql.is_workload_modification),
    Esql.latest_sensitive_rbac_timestamp = max(Esql.sensitive_rbac_timestamp),
    Esql.earliest_workload_modification_timestamp = min(Esql.workload_modification_timestamp),
    Esql.event_action_values = values(event.action),
    Esql.gcp_audit_resource_name_values = values(gcp.audit.resource_name),
    Esql.user_agent_original_values = values(user_agent.original),
    Esql.source_ip_values = values(source.ip)
  by client.user.email
| where Esql.sensitive_rbac_count > 0
    and Esql.workload_modification_count > 0
    and Esql.latest_sensitive_rbac_timestamp <= Esql.earliest_workload_modification_timestamp
| eval Esql.rbac_to_workload_minutes = date_diff(
    "minute",
    Esql.latest_sensitive_rbac_timestamp,
    Esql.earliest_workload_modification_timestamp
  )
| where Esql.rbac_to_workload_minutes <= 5
| keep
    client.user.email,
    Esql.*
'''

[[rule.threat]]
framework = "MITRE ATT&CK"

[[rule.threat.technique]]
id = "T1098"
name = "Account Manipulation"
reference = "https://attack.mitre.org/techniques/T1098/"

[[rule.threat.technique.subtechnique]]
id = "T1098.006"
name = "Additional Container Cluster Roles"
reference = "https://attack.mitre.org/techniques/T1098/006/"

[rule.threat.tactic]
id = "TA0004"
name = "Privilege Escalation"
reference = "https://attack.mitre.org/tactics/TA0004/"

[[rule.threat]]
framework = "MITRE ATT&CK"

[[rule.threat.technique]]
id = "T1098"
name = "Account Manipulation"
reference = "https://attack.mitre.org/techniques/T1098/"

[[rule.threat.technique.subtechnique]]
id = "T1098.006"
name = "Additional Container Cluster Roles"
reference = "https://attack.mitre.org/techniques/T1098/006/"

[rule.threat.tactic]
id = "TA0003"
name = "Persistence"
reference = "https://attack.mitre.org/tactics/TA0003/"

[[ currentRule.title && currentRule.title.length > 60 ? currentRule.title.slice(0, 60) + '…' : currentRule.title ]]

1247ad12-8d6f-4412-84c2-9f42ffbda7df

Propose edits, view related rules, and engage with the community through comments.

Executable code, not a declarative rule

This submission is arbitrary [[ currentRule.format ? currentRule.format.toUpperCase() : '' ]] code, not a declarative detection rule — review it as you would any script before running it, and note that it cannot be safely aggregated or auto-executed alongside YARA/Sigma/Suricata/CRS rules.

Rejected pattern present

[[ reason ]]

What does this mean?

Format & documentation suggestions

Not a security issue — this rule is valid and was accepted as-is. These are optional fields/conventions its own format recommends, mainly affecting its quality score and how well it documents itself for other reviewers.

[[ reason ]]

What do these messages mean?

[[ currentRule.title && currentRule.title.length > 60 ? currentRule.title.slice(0, 60) + '…' : currentRule.title ]]

[[ currentRule.title ]]

Community Discussion