Skip to main content
Back
IntrusionAWS · Google Cloud · Kubernetes · SaaS
16 min read

Zeabur platform intrusion

Victim
Zeabur
When
Oct 14–15, 2026Disclosed Oct 15, 2026
Entry
Leaked long-lived credentials kept in plaintext: a Zeabur AWS access key and a shared cluster admin token with no expiry, used from the public internet. How they were obtained cannot be confirmed.

On October 14 and 15, 2026 an attacker used leaked long-lived credentials, Zeabur AWS keys and a shared cluster admin token with no expiry among them, to read the secrets of every service on the shared cluster (partly confirmed), sign in as root to all 101 Google Cloud Dedicated Servers and read the platform database through a node it added to the internal network. How the credentials were obtained cannot be confirmed; Zeabur's leading hypothesis links them to its August 2026 incident, with no direct evidence. More than 16 hours of activity on October 14 triggered no alert, and on October 17 Zeabur announced that it will shut down on December 31, 2026.

What happened

External Parties
Callback host
GitHub
AI services
+ 5 providers
Internet
Cloud API · long-lived keysusable from the internet
Shared Platform Cluster
Dedicated Servers · Google Cloud
+ secrets service
+ backups
+ platform database
Internal Systems
+ config repo
+ GitHub App key
+ provider tokens
Config Repo + Vendor Tokens
Cluster Admin Interface
Admin Token · no expiry
System Secrets
Tenant Services · 9 regions
Tenant Databases
Build Jobs · AWS keys
SSH Port · internet-facing
Root Passwords
Server Fleet · 101
Added Network Node

Replaying the Zeabur October 2026 platform intrusion

  1. 1

    Standing credentials

    Jan 10, 2024 – Sep 2, 2026

    A cluster admin token with no expiry and AWS keys sit in plaintext in the configuration repository; who downloaded the repository is unconfirmed.

  2. 2

    Cloud key, then cluster admin

    Oct 14 01:36–01:51 UTC

    An AWS key is accepted from the internet and the admin token four minutes later, over the public admin interface; some system secrets are read by a path the report does not state, and a program that calls out is planted, its entry point unconfirmed.

  3. 3

    Every tenant on the shared cluster

    Oct 14 12:39–17:04 UTC

    The settings and secrets of all services across 9 regions are read, partly confirmed; read commands run in 865 tenant database services, an assessed count.

  4. 4

    Dedicated Servers on Google Cloud

    Oct 14 13:Oct 40 – 15 20:36 UTC

    Root sign-ins reach all 101 Dedicated Servers over public SSH, with root passwords the report assesses were read from initialization settings; SSH keys are added and customer disks copied.

  5. 5

    Secrets, backups and the platform database

    Oct 15 14:52–20:30 UTC

    About 53 production secrets are read with the secrets management service key and a customer backup through a presigned URL; a node added to the internal network reads the platform database.

Impact

The settings and secrets of every service on the shared cluster read across 9 regions (partly confirmed), read commands in 865 tenant database services (an assessed count), root sign-ins to all 101 Google Cloud Dedicated Servers, about 53 production secrets read and a customer database backup downloaded, and 3,218,965 environment-variable records and 321,810 accounts read from the platform database; Zeabur announced that it will shut down on December 31, 2026.

Outcome

Where cloud-provider controls fit

On AWS, a service control policy would have stopped new long-lived access keys from being issued, a second one with network conditions would have limited permission-requiring calls from outside your networks if its account is a member account of an organization. The first API call is not named, so refusal of that use is not established, and a resource control policy would have refused the read of the secrets from outside those networks. An EKS private endpoint bounds access to the Kubernetes API, and on GKE an organization policy would have kept a new cluster's endpoint private; the report names neither the shared cluster's cloud nor the admin-interface type, so these settings cannot be credited with refusing the reported admin access. On Google Cloud, a hierarchical firewall policy would have kept SSH closed to the internet, a service perimeter would have refused the read of the server settings with the old key from outside it, OS Login would have ignored the added SSH keys in metadata, though a key that holds an OS Login role still signs in, and an IAM deny policy would have refused the disk copies unless the key belonged to an approved administrator. The cluster token, the reads inside the shared cluster, the self-managed internal network and the database account sit outside the cloud control plane: you own tenant isolation, token expiry and keeping secrets out of plaintext.

Could this have been prevented?

In hindsight, not as a whole. 4 controls would have refused a step outright, and 35 controls would have limited one. 11 steps have no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.

4
Prevented
35
Contained
11
Not covered

How it could have been prevented

The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.

External Parties
Callback host
GitHub
AI services
+ 5 providers
Internet
Cloud API · long-lived keysusable from the internet
Shared Platform Cluster
Dedicated Servers · Google Cloud
+ secrets service
+ backups
+ platform database
Internal Systems
+ config repo
+ GitHub App key
+ provider tokens
Config Repo + Vendor Tokens
Cluster Admin Interface
Admin Token · no expiry
System Secrets
Tenant Services · 9 regions
Tenant Databases
Build Jobs · AWS keys
SSH Port · internet-facing
Root Passwords
Server Fleet · 101
Added Network Node

Replaying the Zeabur October 2026 platform intrusion

1 · Standing credentials

5 Contained2 Not covered2 on other clouds

2 · Cloud key, then cluster admin

10 Contained2 Not covered6 on other clouds

3 · Every tenant on the shared cluster

2 Contained2 Not covered2 on other clouds

4 · Dedicated Servers on Google Cloud

2 Prevented10 Contained1 Not covered8 on other clouds

5 · Secrets, backups and the platform database

2 Prevented8 Contained4 Not covered6 on other clouds

Key facts

  • The shared cluster admin token was created on January 10, 2024 with no expiry and was never rotated; it was written in plaintext to the private configuration repository on January 20, 2026, and the August 2026 rotation covered neither it nor the AWS keys
  • A Zeabur AWS key was first used at 01:36 UTC on October 14 and the admin token at 01:40, over a cluster admin interface open to the public internet; more than 16 hours of activity that day triggered no alert
  • Read commands ran in 865 tenant database services, an assessed count; the database logs recorded commands, not output, so how much data was read cannot be confirmed
  • Root sign-ins were confirmed on all 101 Google Cloud Dedicated Servers of 92 users; SSH keys were added to 50 of them and 16 copies of customer disks were attached to 5 attacker virtual machines
  • On October 15 about 53 production secrets were read with the secrets management service key; through a node added to the internal network, an account that could read all data read 3,218,965 environment-variable records and 321,810 accounts from the platform database
  • The investigation started at 13:56 UTC on October 15 after a user report; the last attacker activity was observed at 21:05, and on October 17 Zeabur announced that it will shut down on December 31, 2026

Sources

Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.

  1. 1.
    Zeabur October 2026 security incident (preliminary report)Zeabur · Oct 19, 2026 · Vendor reportTimeline, impact counts, root causes, attacker steps, containment, open questions, indicators of compromise, glossary
  2. 2.
    Unauthorized Access to Zeabur Infrastructure and User DataZeabur (status page) · Oct 20, 2026 · Status pageDisclosure timeline, initial scope statement, containment, service impact, user guidance
  3. 3.
    Zeabur August 2026 security incidentZeabur · Sep 7, 2026 · Vendor reportAugust incident, prior rotation, prior guidance
  4. 4.
    Zeabur is shutting down: what happens nextZeabur · Oct 17, 2026 · Vendor reportShutdown dates, reasons, migration guidance
  5. 5.
    Zeabur service migration, refunds, and security incident claimsZeabur · Oct 22, 2026 · Vendor reportClaims deadline, data deletion date, compensation process
  6. 6.
    The Shared Cluster Will No Longer Accept New ProjectsZeabur · Feb 23, 2026 · Vendor reportShared cluster retirement
  7. 7.
    Zeabur shutting down after second security breach in two monthsBleepingComputer · Oct 20, 2026 · NewsShutdown reporting, August link as reported, rounded counts
  8. 8.
    Zeabur says leaked plaintext credentials exposed 321,810 accounts, announces shutdownteiss · Oct 21, 2026 · NewsShutdown reporting, admitted failures as reported
  9. 9.
    Zeabur Breach Exposes 321,810 Accounts: What Customers Should Do NoweSecurity Planet · Oct 23, 2026 · NewsCustomer guidance, account data fields as reported, shutdown dates

The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.

More cloud attack breakdowns

All breakdowns