Security update · 1 November 2026
Northflank says customers reported unauthorized use of AI API keys stored on its platform. Its October 30 incident notice advises treating environment variables as compromised, rotating credentials and reviewing provider billing. By October 31, rotation support included RabbitMQ and Minio; audit logs were also available to free users. The latest update checked was October 31 at 23:31 UTC. Northflank’s incident notice.
Why this matters
An API key is an application credential, not just a setting. Replacing text in a deployment does not itself invalidate the old credential at the service that issued it. For administrators, recovery has two distinct goals: stop the old credential working and confirm that legitimate applications can use its replacement. A successful deployment alone does not demonstrate both.
Do not miss inherited secrets
Northflank’s documentation explains that secret groups supply values to services and jobs. Values set directly on a workload override matching group values. Changing a group is therefore not enough to prove every consumer received the intended credential. Check overrides and restart dependent services or run new jobs to apply updated runtime values. Secret-group documentation.
Practical check: keep a restricted inventory of credential names, issuing services, dependent applications and verification status. Do not put secret values in the checklist. Give each dependency an owner; test a normal authorized operation after the change. Treat these as operational recommendations, not evidence that any particular customer was affected.
Use logs to investigate, not to declare certainty
Northflank audit events include the actor, source, time and affected resources; filters help narrow a review to a project or period. They describe activity on Northflank, so review the external AI provider’s records separately. An empty search is not proof that a credential was never used elsewhere. Audit-log documentation.
Follow the incident notice for current instructions. This report does not establish a root cause, identify an attacker or claim that every customer experienced misuse.