How the attack works
Trusted Access is the way Azure services reach the Kubernetes API server of an Azure Kubernetes Service (AKS) cluster through the Azure back end. A role binding on the cluster names a source Azure resource and one or more Trusted Access roles. From then on the system-assigned managed identity of that resource can call the cluster's API server through the AKS regional gateway with the Kubernetes permissions of those roles. Microsoft defines the roles per service, for example for Azure Machine Learning, AKS backup and the agentless container posture of Microsoft Defender for Cloud, and documents that Trusted Access reaches clusters with different configurations, among them private clusters, clusters with local accounts turned off, Microsoft Entra clusters and clusters with authorized IP ranges.
The binding is an Azure resource, created with Microsoft.ContainerService/managedClusters/trustedAccessRoleBindings/write. It is not a Kubernetes role binding, so it does not appear among the RoleBinding and ClusterRoleBinding objects that a cluster review looks at, and the protections that keep people away from the API server do not apply to it. An attacker with write access to a cluster in Azure Resource Manager can bind a resource they control and so give its identity standing access inside the cluster. The Azure Activity log records the write with its caller. Microsoft notes that a new binding can take up to five minutes to take effect. No public incident report was found that describes this at the level of the audit event, so the page rests on what Microsoft documents the feature does.
What an attacker needs
- An Azure role with
Microsoft.ContainerService/managedClusters/trustedAccessRoleBindings/writeon the cluster. Owner and Contributor include it. - A source resource of a type that has Trusted Access roles and a system-assigned managed identity, which the attacker controls or can create.
- A free slot. Microsoft documents a limit of 8 Trusted Access role bindings per cluster.
Impact
The source resource's identity keeps the Kubernetes permissions of the bound roles in the cluster for as long as the binding exists. The access does not depend on a kubeconfig, on network reachability of the API server or on the attacker's own Azure role, and it is granted outside the Kubernetes RBAC objects that cluster administrators review. What it allows depends on the roles in the binding.
Detection rules
AKS Trusted Access Role Binding Created or Changed
Detects a successful create or update of a Trusted Access role binding on an Azure Kubernetes Service cluster (Microsoft.ContainerService/managedClusters/trustedAccessRoleBindings/write). The binding gives the system-assigned managed identity of a source Azure resource the Kubernetes permissions of the bound Trusted Access roles in the cluster, through the Azure back end and regardless of private cluster, local account and authorized IP range settings. It is an Azure resource and not a Kubernetes role binding. The event does not say which source resource and roles the binding names, so read the binding on the cluster after the rule fires. Clusters have at most 8 bindings and they change rarely.
title: AKS Trusted Access Role Binding Created or Changed
id: c10cbaa2-ea87-479e-b747-deea434ce255
status: experimental
description: >-
Detects a successful create or update of a Trusted Access role binding on an
Azure Kubernetes Service cluster
(Microsoft.ContainerService/managedClusters/trustedAccessRoleBindings/write).
The binding gives the system-assigned managed identity of a source Azure
resource the Kubernetes permissions of the bound Trusted Access roles in the
cluster, through the Azure back end and regardless of private cluster,
local account and authorized IP range settings. It is an Azure resource and
not a Kubernetes role binding. The event does not say which source resource
and roles the binding names, so read the binding on the cluster after the
rule fires. Clusters have at most 8 bindings and they change rarely.
references:
- https://attack.mitre.org/techniques/T1098/006/
- https://learn.microsoft.com/en-us/azure/aks/trusted-access-feature
- https://learn.microsoft.com/en-us/azure/role-based-access-control/permissions/containers
author: Forestall
date: 2026-11-18
tags:
- attack.persistence
- attack.privilege-escalation
- attack.t1098.006
logsource:
product: azure
service: activitylogs
detection:
selection:
OperationNameValue: MICROSOFT.CONTAINERSERVICE/MANAGEDCLUSTERS/TRUSTEDACCESSROLEBINDINGS/WRITE
ActivityStatusValue:
- Success
- Succeeded
condition: selection
falsepositives:
- Setup of services that work inside the cluster, such as AKS backup, Azure Machine Learning and Microsoft Defender for Cloud, which create a binding when they are enabled. Match the event to that deployment and exclude the deployment identities in a filter_ selection by Caller once confirmed.
level: medium
False positives
- Setup of services that work inside the cluster, such as AKS backup, Azure Machine Learning and Microsoft Defender for Cloud, which create a binding when they are enabled. Match the event to that deployment and exclude the deployment identities in a filter_ selection by Caller once confirmed.
Triage
- Identify the Caller, the caller's IP address and the cluster. The name of the binding is in the resource ID of the event.
- Read the binding on the cluster. Its properties name the source resource and the roles.
- Check whether the source resource belongs to a service the cluster is meant to use, who created it and when, and who controls it.
- Look for a matching deployment, such as enabling AKS backup, attaching the cluster to a Machine Learning workspace or turning on the agentless container posture of Defender for Cloud, which create bindings as part of their setup.
- If the kube-audit logs of the cluster are collected, review API server requests after the binding was created.
Response
Once the attack is confirmed:
- Delete the role binding from the cluster.
- Contain the caller. Revoke its sessions, rotate its credentials and remove the role that allowed the write.
- Review the source resource. If the attacker created it, delete it, and if it is a legitimate resource, find out who can run code as its identity.
- Review the cluster for what the bound roles allowed, such as new workloads, role bindings, service accounts and reads of Secrets, and remove what cannot be explained.
- Rotate secrets stored in the cluster if the bound roles could read them.
Mitigation
- Limit Owner, Contributor and custom roles with the Trusted Access write action on clusters, and assign them through Privileged Identity Management.
- Keep an inventory of the Trusted Access role bindings each cluster should have, such as the ones for backup and Defender for Cloud, and compare it with the bindings that exist.
- Use Azure Policy to audit or deny Trusted Access role bindings on clusters that should have none.
- Include Trusted Access role bindings in access reviews of a cluster, next to Kubernetes role bindings and Azure role assignments.
- Send the Azure Activity log to your SIEM, so a new binding is seen when it is created.
Frequently asked questions
Is a Trusted Access role binding visible in Kubernetes RBAC?
No. It is an Azure resource of the cluster (`Microsoft.ContainerService/managedClusters/trustedAccessRoleBindings`), not a Kubernetes RoleBinding or ClusterRoleBinding. It is listed through Azure, for example with the Azure CLI command that lists the Trusted Access role bindings of a cluster. A review that only reads the role bindings inside the cluster does not see it.
Does a private cluster or disabling local accounts prevent this access?
No. Microsoft documents that Trusted Access can reach clusters with different configurations, including private clusters, clusters that have local accounts turned off, Microsoft Entra clusters and authorized IP range clusters. The feature exists to give Azure services access in exactly those cases. The control is who may write role bindings on the cluster in Azure.
Which legitimate actions create these bindings?
Services that work inside a cluster set them up, for example AKS backup, Azure Machine Learning and the agentless container posture of Microsoft Defender for Cloud. A cluster normally has a small, stable set of bindings that belong to such services. A binding with an unfamiliar source resource, or one created outside the setup of a service, is what the rule is for.
Is there a public incident that used Trusted Access this way?
No public incident report was found that describes this at the level of the audit event. The page is based on what Microsoft documents about the feature, which is that a role binding gives the managed identity of the source resource the Kubernetes permissions of the bound roles.