Join our Newsletter — 33% off our NHI Course
Home› NHI Breaches› Zeabur Second Breach 2026: How Unrotated Admin Tokens…
Breach analysis Incident: 19 Oct 2026

Zeabur Second Breach 2026: How Unrotated Admin Tokens Exposed 3.2 Million Environment Variables

PDF for newsletter subscribers · free to read online

← All NHI breaches
By NHI Mgmt Group, edited by Lalit Choda Updated 26 October 2026 12 min read
On this page

On 19 October 2026, the cloud deployment platform Zeabur published a preliminary report on its second breach in two months. Between 14 and 15 October 2026, an attacker used internal Zeabur credentials to reach the shared hosting cluster in nine regions, some customer dedicated servers and internal systems, and then read the platform database. Zeabur says about 3.2 million service environment variable records and profile data for about 321,810 accounts were taken. The root cause was an identity failure: long-lived cluster admin tokens with no expiry, stored in plaintext and reachable from the internet, were missed by the credential rotation Zeabur carried out after its August 2026 breach. How the attacker first got hold of them is still unconfirmed. Zeabur has since announced that it will shut down on 31 December 2026. Customers are told to treat every credential ever stored on the platform as exposed.

Key takeaways

  • Zeabur says an attacker read about 3.2 million environment variable records and data on about 321,810 accounts on 14 and 15 October 2026, after reaching its shared cluster in nine regions and the platform database.
  • The keys to the platform were long-lived cluster admin tokens with no expiry, held in plaintext in a configuration repository and usable from the public internet. BleepingComputer reports they had not been rotated since January 2024.
  • Zeabur admits its August report was wrong to say all internal admin keys had been disabled or rotated: "That statement was incomplete, and we are correcting it here."
  • Zeabur confirmed root access to 101 GCP dedicated servers, commands run inside 865 user database services, and more than 16 hours of attacker activity with no alert. The company will close on 31 December 2026.
  • The identity lesson: post-breach rotation is only as good as the inventory behind it. A single forgotten, non-expiring admin token turned one breach into two.

At a glance

OrganisationZeabur, a cloud deployment platform, and its customers (about 321,810 accounts)
WhenFirst suspicious activity 10 October 2026; main intrusion 14 to 15 October 2026; preliminary report published 19 October 2026
AttackerUnknown. Zeabur has not named an actor; its leading hypothesis links the intrusion to credentials leaked in the August 2026 breach, which is not proven
Entry pointInternal credentials: a Zeabur AWS key and long-lived cluster admin tokens stored in plaintext and usable from the internet
Identities abusedNon-expiring cluster admin tokens, an over-broad AWS access key, the GCP default service account, plaintext server root passwords, then customers' stored API keys, cloud keys and database passwords
ImpactAbout 3.2 million environment variable records and 321,810 account profiles read; about 53 production secrets read; 101 GCP servers accessed as root; 865 user database services entered; company shutting down
CategoryNHI. Incident class: confirmed NHI breach (unrotated admin tokens used to take platform-wide secrets)

What happened

Zeabur lets developers deploy applications and databases onto shared clusters or dedicated servers, with configuration and secrets held as environment variables. In late August 2026 it was breached for the first time. According to its report of 7 September, an attacker exploited a remote code execution flaw (CVE-2026-7644) in a customer's NextChat service on a shared cluster, then moved through a routing cache holding Legacy API Keys, a long-lived AWS root key in an internal service and an over-privileged database credential to reach production. "The attack began with a vulnerability in a user's publicly accessible service and escalated through four system boundaries," Zeabur wrote. It said it had disabled the Legacy API Keys and rotated internal admin keys, database passwords and cache passwords.

That rotation missed something. Zeabur's October report, written by Yuanlin Lin and published on 19 October 2026, says the first anonymous API activity from an IP address later tied to credential abuse appeared on 10 October, followed by failed SSH sign-ins against three dedicated servers on 13 October. At 01:36 UTC on 14 October a Zeabur AWS key was used, and four minutes later an admin token gave the attacker access to the shared cluster. By 12:39 the settings and secrets of every service on the shared cluster had been read. From 13:40 the attacker ran commands inside customer database containers and began signing in to GCP dedicated servers. On 15 October, after adding a node to Zeabur's internal network using admin credentials obtained earlier, the attacker read the platform database between 16:10 and 16:47 UTC.

Zeabur only began investigating at 13:56 UTC on 15 October, after a user report, and says more than 16 hours of attacker activity triggered no alerts. It removed the attacker's network node at 17:58, rotated the login signing key and signed out every user at 20:25, and revoked Hetzner and DigitalOcean tokens early on 16 October. The report is frank about the cause: "long-lived admin credentials were stored in plaintext and could be used from the public internet." It also corrects its earlier account. The August report had said "all internal admin keys have been disabled or rotated", and Zeabur now writes: "That statement was incomplete, and we are correcting it here."

BleepingComputer, reporting on 20 October, added that the admin tokens had not been rotated since January 2024, that administrator portals were reachable from the internet, and that the attacker added SSH keys to 50 servers and attached 16 customer disk copies to their own virtual machines. Zeabur's report says the attacker searched the stolen data mainly for keys it could use straight away, especially AI service keys, and tried to call Amazon Bedrock with Zeabur's own AWS keys, without success. Zeabur says it "has announced that it will shut down on December 31, 2026", describing it as an operational decision; BleepingComputer reports renewals stop and shared cluster services pause on 1 December 2026.

Timeline

DateEvent
27 August 2026First Zeabur breach begins through a vulnerable customer NextChat service.
7 September 2026Zeabur publishes its August incident report, stating internal admin keys were disabled or rotated.
10 October 2026First anonymous API activity from an IP address later tied to credential abuse.
13 October 2026Failed SSH sign-in attempts against three dedicated servers.
14 October 2026A Zeabur AWS key and an admin token are used; secrets of every shared cluster service are read; commands run in customer databases.
15 October 2026Investigation starts after a user report; the attacker reads the platform database; Zeabur rotates its login signing key and signs out all users.
16 October 2026Hetzner and DigitalOcean tokens revoked.
19 October 2026Zeabur publishes its preliminary October report and confirms it will shut down.
20 October 2026BleepingComputer reports the second breach and the shutdown plan.
31 December 2026Planned closure of the Zeabur platform.

How it happened: the identity attack path

  1. Credentials obtained. The attacker held a Zeabur AWS key and internal admin credentials. Zeabur's leading hypothesis is that they trace back to a GitHub App private key leaked in the August breach, but it has not confirmed this.
  2. Non-expiring admin token used. Long-lived cluster admin tokens, kept in plaintext in a configuration repository and usable from the public internet, gave direct control of the shared cluster. They had no expiry and were left out of the August rotation.
  3. Secrets read across the cluster. With admin rights the attacker read the settings and secrets of every shared cluster service and ran commands inside customer database containers.
  4. Wider cloud access through over-broad identities. An AWS key that could read every bucket in its account, a GCP default service account with project-level write access and plaintext root passwords for GCP servers opened backups and customer servers. Backups held all secrets, with the backup key stored beside them.
  5. Platform database reached. A node added to the internal network with admin credentials could reach the platform database, which was read in full, giving about 3.2 million environment variable records.

Impact

  • Confirmed by Zeabur: about 3.2 million environment variable records and about 321,810 account profiles read; about 53 production secrets read; 566 zSend API keys and 11,536 platform access tokens (stored as hashes) revoked; root access to 101 GCP dedicated servers belonging to 92 users, according to BleepingComputer.
  • Assessed but not fully confirmed: Zeabur assesses that data was read from 865 user database services where commands were run. Whether some database backups in two regions were downloaded is unconfirmed.
  • Possible misuse: Zeabur suspects an OpenAI key stored by a user was used for about 1,107 image generation requests, but says this is not proven. Attempts to use Zeabur's AWS keys against Amazon Bedrock failed.
  • Customers: every credential held in a Zeabur environment variable, from AI provider keys to cloud and database credentials, must be treated as exposed. Projects that moved variables to their own servers after August were outside the database read, though compromised hosts could still expose them.
  • Business: Zeabur will shut down on 31 December 2026, citing operational reasons, and its customers must migrate.

What this means for NHI and AI agent security

This is an NHI breach from start to finish. The attacker used machine credentials: an admin token, a cloud key, a default service account. They took machine credentials in return, millions of them, belonging to other companies. Deployment platforms are attractive for exactly this reason. They concentrate the API keys, database passwords and cloud keys of thousands of customers behind a small set of highly privileged internal identities, and the October wave sits alongside the Railway breach 2026 and the OpenCode Console breach 2026, where stored AI provider keys were also the prize.

The deeper lesson is about incident response. After August, Zeabur said its internal admin keys had been rotated. It could only rotate what it knew about, and tokens that had sat untouched since January 2024 with no expiry were not on the list. This is the risk described in the OWASP NHI Top 10 NHI7 Long-Lived Secrets Guide: a credential that never expires stays valid long after everyone has forgotten it exists. Rotation after a breach has to start from a complete inventory of privileged identities, and it has to be verified, not declared.

For customers, the case shows that secrets handed to a platform inherit the platform's weakest identity. Teams that keep AI provider keys in third-party environment variables should assume they can leak and limit what each key can spend, as our LLMjacking Guide describes.

Recommendations

  • Rotate every secret stored on Zeabur now. Revoke old API keys, cloud keys, database passwords and signing keys before issuing new ones, then plan migration before 31 December 2026. See our Secrets Rotation Runbook Template.
  • Inventory privileged machine identities before declaring rotation complete. List every admin token, cloud key and service account, including those in configuration repositories, and confirm each one was rotated. See our Non-Human Identity (NHI) Guide.
  • Give admin tokens an expiry. Replace static, non-expiring cluster credentials with short-lived ones issued on demand, and block their use from the public internet.
  • Keep secrets out of plaintext configuration. Store admin credentials, root passwords and backup keys in a secrets manager, and never keep a backup key beside the backup. See our Secrets Management Guide.
  • Cut back default and over-broad cloud identities. Scope storage keys to the buckets they need and remove project-wide write rights from default service accounts.
  • Alert on privileged identity use. Admin token use from new locations, new nodes joining internal networks and bulk secret reads should page someone within minutes, not 16 hours.
  • Limit the blast radius of AI keys you hand to platforms. Set spending caps and usage alerts with your AI providers and use separate keys per platform. See our LLMjacking Guide.

Frequently asked questions

What caused Zeabur's second breach in October 2026?

Zeabur says the main cause was long-lived cluster admin tokens with no expiry, stored in plaintext and usable from the public internet, which were not rotated after its August 2026 breach. Over-broad cloud permissions, plaintext server root passwords, weak isolation of the shared cluster and missing alerts made it worse. How the attacker first obtained the credentials has not been confirmed.

What data was stolen in the Zeabur breach?

According to Zeabur, about 3.2 million service environment variable records and profile data for about 321,810 accounts were read from the platform database. Environment variables often hold API keys, AI provider keys, cloud credentials and database passwords. The attacker also gained root access to 101 GCP dedicated servers and ran commands inside 865 customer database services.

Is Zeabur shutting down, and what should customers do?

Yes. Zeabur has announced it will shut down on 31 December 2026, with renewals and shared cluster services paused from 1 December, according to BleepingComputer. Customers should rotate every credential ever stored on Zeabur, sign in again and recreate access tokens, check bills and GitHub audit logs, rebuild affected servers and migrate before the closure date.

Railway breach 2026 · OpenCode Console breach 2026 · LLMjacking 2024 to 2026 · OWASP NHI Top 10 NHI7 Long-Lived Secrets Guide · Secrets Rotation Runbook Template

How NHI Mgmt Group can help

A breach is rarely over when the first round of rotation is announced. We help platform and security teams build a full inventory of privileged machine identities, replace static admin tokens with short-lived credentials and prove that post-incident rotation actually covered everything. See our NHI and AI agent security training.

References

PDF for subscribers

Take this breach analysis with you

The full analysis in one PDF: timeline, attack path, impact and recommendations, with links to the sources and related pages. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 26 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.