Northflank security breach leads to abuse of customers' AI API keys

What it is

Northflank, a cloud platform for deploying applications, containers, databases and AI workloads, disclosed a breach on 30 October 2026. Wiz alerted the company at 11:43 UTC, and engineers reviewed the alert a minute later. Customers then reported unauthorized use of AI API keys stored on the platform. Northflank's guidance was: "Assume all environment variables have been decrypted. Disable/rotate all keys."

Preconditions / Affected

Any secret stored in Northflank is in scope: workload environment variables, secret groups and global secrets. Database add-ons with public access enabled are also a concern. Northflank has not said how many customers were affected or how attackers got in. It has not said whether credentials other than AI API keys were taken.

What to do

Northflank's stated steps for customers:

• Rotate every user-provided secret in workloads, secret groups and global secrets, not only AI keys.
• Review AI provider usage and billing logs for unexpected activity. Report suspicious use to Northflank and to the AI provider.
• Disable public access on add-ons where it is enabled.
• Use the new credential rotation for PostgreSQL, MongoDB, MySQL and Redis add-ons.
• Use the audit logs, now enabled for all customers including free accounts, to investigate suspicious activity.

Because the same environment variables often hold database passwords, cloud credentials and tokens, rotation should also cover credentials for downstream services, not just model providers.

Caveats

Northflank's timeline shows a network policy patch, then further patches at 12:50 UTC and 13:05 UTC. It does not explain the root cause. BleepingComputer's questions to Northflank were unanswered at publication.

Wiz's "Black October for AI Developer Platforms" report counts at least nine incidents in the month, including Railway, OpenCode, Warp and Lightning AI. Railway also involved possible exposure of decrypted environment variables. OpenCode exposed AI API keys. Warp exposed AI agent logs and user telemetry. Wiz has not confirmed a single attacker behind all of them. Northflank says its customer reports "may be tied" to the attacks Wiz saw, which is not a confirmed link.

In two Wiz cases, stolen enterprise credentials drove about 2.5 million Amazon Bedrock calls, costing about $100,000. Wiz found infrastructure that resells model access but could not tie it to the initial breaches.