Northflank Says Attackers Decrypted Customer Secrets About Nine Hours Before Its First Security Alert

Northflank now has a timeline, and it is not a comfortable one. The cloud hosting platform says the first known decryption of customer secrets took place between 02:39 and 03:08 UTC on October 30. Its first security alert arrived at 11:43 UTC, roughly nine hours later, according to the latest update on its incident page.

Northflank lets developers run apps, databases and AI services without managing servers, which means its customers store the keys those apps need: database passwords, cloud credentials, payment keys and, above all lately, API keys for AI providers. On the day of the breach the company told every customer to assume all of their environment variables had been decrypted and to disable and rotate all keys. Some customers have since reported unauthorized use of their AI keys.

What Northflank says now

The company says it patched the exploit within an hour of the alert, which came from the cloud security firm Wiz, and notified customers within two hours. It found no evidence that running workloads or their data were accessed, and says the attack focused on secrets held in its control plane, the part of the platform that manages customer projects. Customers on dedicated or forward-deployed control planes were not affected.

Where it could, Northflank revoked its own access tokens, sessions, code repository integrations and cloud roles on customers' behalf. Independent forensic auditors are still reviewing the attack, and the company has promised a root cause analysis. It has not said how the attacker got in or how many customers lost secrets.

Part of a much bigger month

Northflank was the last name on a list Wiz published the same day, under the title Black October. It counts at least nine AI developer platforms that disclosed incidents during October, including Fireworks AI, Lightning AI, OpenCode, Warp, Zeabur and Railway, which told its users to rotate every key a week earlier. Zeabur is shutting down because of its breach.

The list keeps growing. Wiz has since added Trigger.dev, Chainstack, kernel.sh and Blacksmith. It has not found a single vulnerability linking the attacks, but the target is always the same: stored environment variables, especially the ones that unlock AI models. Stolen keys are resold or used directly, a practice known as LLMjacking. In two enterprise cases Wiz traced, stolen Amazon Bedrock credentials made well over two million model calls and ran up more than $100,000 in usage.

The part customers still have to do

Changing a secret inside Northflank does not kill the stolen copy. The old key keeps working until it is revoked at the provider that issued it, whether that is OpenAI, Anthropic, AWS or a payment processor. Northflank's guidance is to rotate integration and project secrets at the source, check provider logs and bills since October 30, and turn on IP restrictions and spending limits wherever a provider offers them.

Some offers on this page may be paid placements or contain affiliate links.