Northflank Breach: Why AI API Keys Are Now a Target

The Northflank breach is a security incident that cloud hosting platform Northflank disclosed on October 30, 2026, telling customers to treat every secret in their environment variables as stolen. Multiple customers reported unauthorized use of AI API keys stored on the platform. According to security firm Wiz, it is not an isolated case: at least nine AI developer platforms disclosed incidents in October. The lesson for businesses is plain. An AI key is now a secret with direct monetary value. If one sits on such a platform, revoke and replace it at the provider first, then review usage and billing logs.

What changed with the Northflank breach on October 30?

On its status page, Northflank told customers on October 30, 2026 to rotate environment variables immediately, after a threat alert from security firm Wiz. A second note the same day was blunter: assume all environment variables have been decrypted, and disable or rotate all keys. That day, multiple users reported that AI keys stored on the platform had been used by someone else. According to BleepingComputer, Northflank has not said how the attackers got in or how many customers were affected.

Northflank's November 2 update narrows the picture somewhat. The company says the attack vector was contained within an hour of the first alert, that running workloads and their data show no sign of access, and that the attack focused on secrets in the control plane. Independent forensic auditors are reviewing the incident, and a root cause analysis will follow once findings are confirmed. In the meantime, the platform enabled secret rotation for Postgres, MongoDB, MySQL, Redis, MinIO and RabbitMQ add-ons and turned on audit logs for all customers, including free accounts.

Black October: which AI platforms were hit in October?

Black October is Wiz's name for the wave of attacks during the month. In its October 30 report, Wiz writes that at least nine AI developer platforms disclosed incidents: Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode and Northflank. The report was later updated with Brev (NVIDIA) and Trigger.dev. Zeabur is shutting down after its incident.

According to Wiz, no single vulnerability or dependency links these attacks, but the target is shared: attackers are hunting for stored environment variables, especially those that grant AI access. Fireworks, Zeabur and Lightning attribute the root cause to credential access. Railway traces initial access to privilege escalation in its build infrastructure, while Warp points to a third-party vendor dependency.

What does a stolen AI API key cost a business?

A stolen AI API key is a model account whose bill lands on you. This abuse is called LLMjacking: using stolen credentials to run models on someone else's account, or reselling that access through brokers. In two enterprise cases Wiz investigated, stolen credentials were used on Amazon Bedrock. The first case saw about 800,000 calls over more than 30 hours, the second about 1.7 million calls over more than 13 hours. At list price, Wiz says, the usage in these two cases exceeded $100,000. In the first case, about three quarters of successful calls went to Anthropic's Opus models.

Timing matters too. In the first case, the key was tested within 30 minutes of being stolen. In the second, the heavy traffic surfaced seven days after the first test. Our reading: seeing nothing unusual on today's bill does not prove a key is safe. Wiz also mentions consumer bills in the hundreds to thousands of dollars and an alleged theft of about $371,000 using crypto keys stored in Railway.

What did not change with the Northflank breach?

Many questions about the Northflank breach are still open, and the basic rule of key security stays the same. Four points need separating.

Entry point unknown: Northflank has not yet disclosed how the attacker got in or how many customers were affected. A root cause analysis is pending.
Still under analysis: Wiz writes that most of these incidents are still being analyzed and detailed technical information is not available.
Resale link not proven: Wiz states plainly that it cannot definitively tie the brokers reselling stolen keys to these breaches. Its report does, however, point to an active market for such keys.
The risk sits with your key: Because no single flaw links the attacks, "we don't use that platform" is not enough. Every third-party service holding your keys raises the same question.

Which keys should you rotate first?

For the rotation order, Northflank's November 2 guidance is a good starting point: AI provider keys first, then cloud credentials, payment keys, private keys and database credentials. The table below turns that order into business terms. The "What happens if stolen?" column is our own assessment.

Secret type	What happens if stolen?	First action
AI provider key	Someone runs models on your account and the bill comes to you.	Revoke the old key in the provider's console, create a new one, check usage and billing.
Cloud credential	Models and resources in your cloud account are used in your name. Both Wiz cases ran on Amazon Bedrock.	Revoke the credential and review access logs.
Payment key	Transactions can be made on your payment account in your name.	Replace the key at the payment provider and check recent transactions.
Private key	Signing or wallet access passes to someone else. Wiz reports an alleged theft of $371,000 using crypto keys stored in Railway.	Replace the key and move linked accounts to the new one.
Database credential	Customer records become reachable, and a personal data breach comes into question.	Change the password, review access logs, assess whether it is a breach.

What should you watch for when rotating keys?

When rotating keys, watch four points: where you revoke, delayed abuse, integrations and legal notification.

Revoke at the provider: Northflank advises not only changing the key in its own dashboard but revoking it at the provider and generating a replacement. As long as the old key stays valid at the provider, the stolen copy keeps working.
Logs and billing: Review your provider's usage and billing logs. Northflank suggests contacting providers such as OpenAI or Anthropic about unexpected usage. Wiz's second case shows abuse can start days later.
Integrations: Northflank revoked its own access tokens, sessions and version control integrations. Customers need to sign in again and reconnect their integrations.
GDPR: If your environment variables also held a customer database password, assess whether this is a personal data breach. Under Article 33 GDPR, a controller must notify the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of a breach, unless it is unlikely to result in a risk to individuals. Breaches must also be documented. Northflank says it found no sign of access to running workloads or data, but your assessment should rest on your own logs. This is not legal advice.

What does this news mean for your business?

For most businesses, the Northflank breach is not a reason to switch hosting tomorrow but a prompt to ask where your keys actually live. Take an online store that runs an AI assistant answering order and return questions, built by a small agency on a platform like those in Wiz's list. If the assistant's AI key, the payment provider key and the order database password all sit in the same set of environment variables, one leak means a surprise AI bill, a payment key to replace and a possible breach notification at once. The store owner may not even know which platform hosts the code, so the agency needs to be able to say so in writing.

Who these steps suit, and who they do not:

Suits: Businesses that host apps on Northflank, Railway or another platform on Wiz's list, keep AI, payment or database keys in environment variables, or rely on an agency or freelancer for development.
Does not suit: Businesses that only use ChatGPT or Claude through the web interface and have never created an API key have no keys to rotate.

Three steps you can take this week:

Build a key inventory: which AI, cloud, payment and database keys sit on which platform, with which agency or in which automation tool?
If you use a platform on Wiz's list, revoke and replace keys at the provider in the order of the table above.
Review your AI provider's usage and billing logs since the start of October and report any increase you cannot explain to the provider and the platform.

When you set up your own AI assistant, plan where its keys will live from day one; our AI assistant installation guide is a good place to start. For repeating workflows, see our automation service, and for multi-step work our AI agent service. We looked at where ready-made tools fall short in our no-code automation tools article and at usage limits and credits in our Claude Ultra article.

Sources

Last checked: November 2, 2026.

Next step

To get clear on where and how your AI assistant and automations keep their keys, visit our AI assistant service page, request a short discovery call through our contact form, or call +90 850 840 08 23 and talk live with Azra, our voice assistant.

Frequently asked questions

Am I affected if I don't use Northflank?

Not directly, but the risk is not limited to Northflank. Wiz counts incidents at no fewer than nine AI developer platforms in October, with no single flaw linking them. Check the statements of every platform and agency that holds your keys.

What is LLMjacking?

LLMjacking means using stolen cloud or AI credentials to run AI models on someone else's account, or selling that access. The bill goes to the key's owner. In the two cases Wiz investigated, usage at list price exceeded $100,000.

How can I tell if my AI API key was stolen?

Look for increases you cannot explain in your provider's usage and billing logs. In one Wiz case, the heavy abuse began seven days after the first test, so a clean log today is not proof. If in doubt, revoke and replace the key at the provider.

I rotated the key. Do I need to report a breach?

That depends on whether personal data was accessed without authorization. If your environment variables also held a customer database password, assess the situation. Under Article 33 GDPR, a notifiable breach goes to the supervisory authority within 72 hours where feasible. If unsure, ask your legal counsel.

The Northflank Breach and Black October: What Changes for Your AI API Keys