· via dev.to (home feed)
Rackspace Spot breached via leaked Kubernetes credentials, secrets read in under eight minutes
A dev.to report says attackers used service-account credentials leaked in a public repository to enter Rackspace Spot and read customer secrets less than eight minutes after logging in.

What happened
A report published on dev.to describes a breach of Rackspace Spot, the company's managed Kubernetes platform, that began on November 17, 2026. According to the post, the attackers' way in was a set of service-account credentials that had previously been exposed in a public repository.
The report says the intruders moved quickly once authenticated. Within minutes they were enumerating resources across several customer environments and standing up new pods configured with elevated privileges. Those pods were reportedly named to match ordinary naming conventions so they would not stand out to monitoring, after which they mounted service-account tokens and issued kubectl commands to list and pull secrets from customer namespaces.
The targeted material was high value: database credentials, API keys and TLS certificates. According to the report, dozens of secret objects were taken from each affected cluster before rate limiting or anomaly detection raised an alert, and forensic timelines place the first secret-read operation at less than eight minutes after the initial login.
Why managed platforms widen the blast radius
Rackspace Spot abstracts infrastructure provisioning behind simplified APIs and integrated controls. The dev.to report argues that this convenience concentrates permissions in ways that enlarge the impact of a single credential compromise. Because the stolen tokens carried the same rights as a legitimate user, the attackers could enumerate cluster nodes, inspect attached volumes and walk through namespace configurations without drawing immediate attention, then query the Kubernetes Secrets store directly. The objects there reportedly included plaintext passwords for databases and service accounts along with external API keys, turning one identity compromise into an infrastructure-wide exposure.
From secrets to production systems
The report also maps out how stolen secrets become production incidents. A secret holding a PostgreSQL or MySQL connection string can let an attacker connect straight to the backing managed database, sidestepping network policies that only guard the cluster itself. API keys for internal services such as payment or logging providers open further pivoting opportunities, and service-account tokens found in secrets often carry rights to create pods or jobs in production namespaces, which can be used to mount the same persistent volumes and harvest still more secrets from message queues, caches and object storage.
Manual review, the report contends, cannot track these chained paths in real time, because the bindings that matter — a secret mounted into a newly scheduled pod, or a cross-namespace RoleBinding — often only exist at runtime. It argues instead for automatically correlating secret-access logs with workload identity graphs.
What the report recommends
The dev.to post lays out several mitigations:
- Continuous, event-driven compliance scanning of clusters to catch plaintext secrets in ConfigMaps or environment variables and RBAC roles granting wildcard verbs, checked against hardened baselines derived from CIS Kubernetes benchmarks.
- Pipeline integration so infrastructure-as-code is validated before deployment, plus drift detection to catch changes that reintroduce weak bindings or unencrypted secrets after the fact.
- Layer 7 enforcement at ingress controllers or service meshes — for example, Envoy rules requiring both mTLS identity validation and a matching RBAC annotation before any request to secret endpoints is allowed through, with denied requests logged alongside the source workload identity.
- As immediate steps, auditing all service accounts and removing overly permissive roles.
Why it matters
If the reported timeline holds, the incident is a compact demonstration of how quickly a leaked long-lived credential turns into broad data exposure: under eight minutes from login to the first secret read, with no just-in-time permission boundary in between to interrupt the chain. For teams running Kubernetes anywhere, the takeaway is that perimeter controls say little once a valid token is in hand — the defences that matter are short-lived and narrowly scoped identities, encrypted and isolated secrets, least-privilege RBAC, and runtime signals fast enough to catch secret enumeration while it is happening. For customers of managed platforms, it also raises the familiar shared-responsibility question of who is watching the identities that the platform's own abstractions create.
- #kubernetes
- #cloud-security
- #rackspace
- #secrets-management
- #rbac