CLOUD BREACH

Rackspace Spot Breach

Leaked Credentials Exposed Customer Kubernetes Secrets for Nearly 22 Hours. An attacker used leaked credentials to act as legitimate users inside Rackspace Spot on November 17, 2026.

Matt Lucas  |  November 22, 2026  |  5 min
Editorial hero illustration
~21.7 hrs
Attacker access window on Nov 17
131,072
Addresses in the flagged /15 range
4 days
From intrusion to full service restore
0
Platform alerts: a customer reported it
TL;DR
  • What: On November 17, 2026, an unauthorized party used leaked credentials to get into Rackspace Spot and act as legitimate users in some customer organizations.
  • Impact: The attacker could create pods, read every Kubernetes Secret in affected users' clusters, list nodes and volumes, and access those users' Spot API and refresh tokens.
  • Fix / mitigation: Rackspace revoked tokens and sessions, removed the rogue pods, rotated keys and tightened monitoring, and told affected customers to reissue Spot tokens, refresh kubeconfigs and rotate every credential stored in cluster Secrets.
  • Who's at risk: Rackspace Spot customers, plus every downstream database, registry, Git, storage and SaaS service whose credentials were stored as Secrets in affected clusters.

An unauthorized party used leaked credentials to operate inside Rackspace Spot, the managed Kubernetes platform Rackspace built with Platform9. On November 17, 2026 they had access for about 21 hours and 40 minutes. During that time they acted as legitimate users in some customer organizations: they created pods, read Kubernetes Secrets, and listed nodes and volumes. Rackspace says the attacker could have read any Secret in an affected user's clusters, along with that user's Spot API tokens and refresh tokens.

Rackspace did not catch the intrusion itself. It found out on November 18 from a customer report. The administration console then stayed offline until full service returned on November 21. That meant customers could not use the console to rotate the credentials they had just been told were compromised.

What happened

According to the customer notification obtained by BleepingComputer, the attacker had access from about 00:20 to 22:00 UTC on November 17. They got in with leaked credentials, not by exploiting a software flaw. Rackspace has not said whose credentials leaked, how they leaked, or whether they belonged to a customer, an employee or a service account. It has also not said how many customers were affected or whether any data was taken. BleepingComputer asked and had no response at publication.

The permissions the attacker used come down to four high-value capabilities:

Why Kubernetes Secrets turn one breach into many

To anyone with read access through the API, Kubernetes Secrets are only base64 encoded, not encrypted. Any identity allowed to get or list secrets in a namespace sees the plaintext values. On a managed platform where the attacker holds a legitimate user's session, encryption at rest in etcd does not help, because the API server decrypts Secrets on request.

That makes the blast radius bigger than Rackspace Spot. A typical production cluster stores credentials for a managed database, an object storage bucket, a container registry, a Git provider and several SaaS APIs. Each of those is a separate system where the attacker may now hold valid access. None of them will flag the activity as unusual, because the credential is real. Rackspace's own guidance reflects this: it told customers to review logs for databases, container registries, Git repositories, cloud storage and third-party services going back to November 17.

Detection lag is the core failure

The attacker spent nearly 22 hours on a multi-tenant control plane, and the platform raised no alert. Discovery came from a customer report the following day. Organizations that counted on the provider to spot abuse of its own console got no warning until after the attacker was gone.

Indicators of compromise

Rackspace published the following indicators:

The /15 range is wide. Matching traffic is a lead to investigate, not proof of attacker activity. The name prefixes point more directly at this incident, but short, generic strings like i1 and probe will match legitimate names in some environments. A match needs context before it means anything: was the object created on November 17, by which identity, and from which image source? Rackspace has not explained what the prefixes mean.

Rackspace's response and the outage

Rackspace says it revoked affected tokens and sessions, removed the unauthorized pods, rotated keys, and strengthened access controls and monitoring. It told affected customers to reissue Spot API tokens, replace refresh tokens used in automation tools, download fresh Kubernetes configuration files, and rotate the credentials stored in their cluster Secrets.

The way the response was sequenced caused friction. Rackspace took the console offline during remediation, and restoration slipped several times before it declared full service on November 21. Sirish Raghuram, co-founder and Chief Growth Officer of Platform9, said on Reddit that the shutdown was deliberate: "we made the decision that shutting all access down and sanitizing the entire environment is what we can do to best protect everyone using Spot." He added that the team was "incredibly disappointed to have let you down in this way."

Shutting everything down is a defensible containment choice. It also meant that for several days, customers who knew their Spot tokens were compromised could not reissue them through the console. Credentials for external systems stored in Secrets could still be rotated at those systems, and that is where the larger exposure lies.

What this says about managed Kubernetes risk

This was not a container escape, a kernel bug or a Kubernetes CVE. Someone reached a control plane with valid credentials and then made ordinary API calls. Three points matter for any organization running workloads on a managed Kubernetes service:

Scope of what is known

Rackspace has not released the number of affected customers, the source of the leaked credentials, or whether data was stolen. Which customers count as affected is defined only by Rackspace's notification. An organization that used Spot on November 17 but received no notice has nothing public to rule exposure in or out beyond the published indicators.

RedEye assessment

The credential leak is the headline, but the lasting damage is in the downstream systems whose keys sat in those Secrets. Those systems will record valid logins, not attacks. For Rackspace Spot customers, the realistic assumption is that every Secret in an affected cluster was read, and the real exposure is whatever those credentials could reach. RedEye can review cloud, registry, source control and SaaS activity from November 17 onward and map which downstream credentials carry the most risk.

Questions about your exposure?

RedEye Security provides assessments for organizations that need to understand their real risk.

Talk to us