Cyber Incident Victim: Microsoft
Date:
Jun 2020
Location:
United States of America
Summary
Attackers compromised misconfigured machine-learning clusters within a major cloud provider's Azure service, exploiting exposed Kubeflow dashboards to deploy malicious containers that surreptitiously mined Monero cryptocurrency. The clusters, running Kubernetes-based Kubeflow environments, were vulnerable due to customers overriding default security settings—specifically by enabling direct internet access to administrative dashboards for convenience, bypassing secure ingress controls. This allowed unauthorized actors to execute cryptojacking operations, leveraging high-powered computing resources at customer expense. Security researchers identified tens of affected clusters utilizing a publicly sourced container image containing hidden mining code, marking the first observed campaign specifically targeting Kubeflow infrastructure through this attack vector.
| CIA Posture | Motives | Tactics, Techniques & Procedures |
|---|---|---|
| Available to members | 1 motive | 1 technique |
| Threat Actors | Type | Location |
|---|---|---|
| 0 actors | Available to members | Available to members |
Description
In June 2020, Microsoft disclosed that attackers compromised machine-learning clusters within its Azure cloud service to mine cryptocurrency, exploiting misconfigured customer environments. The incident involved Kubernetes clusters running Kubeflow, an open-source machine-learning framework. Attackers targeted clusters where users had altered default security settings, exposing the Kubeflow administrative dashboard directly to the internet instead of restricting access through the secure istio ingress gateway. This configuration error enabled unauthorized internet access to the dashboard, allowing attackers to deploy malicious containers across the clusters. Microsoft identified tens of affected clusters, many running a publicly available container image that secretly contained cryptocurrency mining code. The attackers leveraged this access to create or modify Jupyter Notebook servers within the clusters, implanting images designed to mine Monero—a privacy-focused cryptocurrency known for being computationally intensive to generate.
Microsoft's Azure Security Center detected the campaign through routine monitoring, marking the first observed attack specifically targeting Kubeflow environments. Security researchers, including engineer Yossi Weizman, traced the compromise to customers exposing the Kubeflow dashboard for convenience, bypassing the default requirement to tunnel access through the Kubernetes API server. This insecure practice allowed anonymous attackers to perform administrative actions, such as deploying unauthorized containers. The cryptojacking operation consumed substantial computational resources rented by Azure customers, effectively transferring mining costs to them while generating illicit revenue. Microsoft notified affected customers and provided detection guidance, including checking for suspicious container deployments and unauthorized dashboard exposures. The company noted this attack shared characteristics with prior campaigns against internet-exposed Kubernetes services but represented a novel focus on machine-learning workloads.