Northflank breach: customers' AI API keys used without authorization

Northflank, a cloud platform used to deploy applications and databases, has disclosed a security incident in which attackers appear to have obtained customer secrets. Several of its users then reported that AI provider API keys they had stored on the platform were being used by someone else.

What Northflank has disclosed

According to the incident timeline relayed by teiss and Cyber Security News, the episode began on October 30 with a threat alert from the security company Wiz at 11:43 UTC. Northflank engineers looked at it within a minute, deployed a first network policy patch at 12:02 and further patches at 12:50 and 13:05, and asked Wiz's incident response team for help. At 13:58 UTC the company told customers to assume that all environment variables had been decrypted and that the secrets in them were stolen. Later that day it said users were reporting unauthorized use of the AI keys they had supplied.

An update published on November 2, as summarized by a DEV Community post that quotes the status page, states that running workloads were not affected. The same post reports that the attack vector was contained within an hour of the first alert, that access tokens, sessions and some integrations were revoked, and that independent forensic auditors are reviewing the case before a root cause analysis is released. The author reads this as pointing to secrets held in Northflank's control plane rather than in customer containers; that is an interpretation, not a confirmed finding.

What customers are asked to do

Northflank advises rotating secrets stored in workloads, secret groups and global settings, disabling public access to add-ons, revoking old credentials with the issuing providers, and checking AI usage and billing logs. The company has enabled audit logs for all accounts, including free ones. It has not said how attackers got in or how many customers were affected, and teiss reports that its questions went unanswered.

A wider pattern

Wiz has published a report titled "Black October for AI Developer Platforms" that, according to Resultsense, counts at least nine incidents at such platforms during October, among them Railway, OpenCode, Warp and Lightning AI. The financial risk is so-called LLMjacking: stolen keys are used to run models at the victim's expense. In two separate cases Wiz investigated, unrelated to Northflank, compromised credentials drove roughly $100,000 of Amazon Bedrock usage.