Observed incident

Second Zeabur breach exposes customer secrets and targets AI keys

Reported

Published here

What happened

Attackers used internal credentials to reach shared clusters and managed servers, then read production secrets and searched customer data for AI-service keys. Zeabur confirmed root sign-ins on 101 GCP servers. All attempts to invoke Bedrock with Zeabur’s AWS keys failed.

Parties and reporting

AI tools and systems: Customer AI-provider API keys, Amazon Bedrock, Hosted n8n and new-api/one-api, Zeabur control plane

Outcome

Zeabur blocked known access, revoked tokens and rotated credentials again after a backup captured a replacement key. On October 28 it said affected customers should pursue third-party charge disputes with each provider.

Operating lesson

Invalidate exposed backups and deployment credentials before issuing replacements. Remove unauthorized SSH keys, check AI billing and save provider logs for disputes.

Evidence and limits

This October intrusion followed the August breach. An August GitHub App key as the source of the October credentials remains a hypothesis. Zeabur distinguishes confirmed reads, missing logs and suspected downstream OpenAI usage. No attacker model was identified.

Sources