Northflank Breach Triggers Abuse of Customer AI Keys
Cloud hosting provider Northflank has advised its users to immediately change all credentials and stored secrets following a security incident that resulted in the exploitation of customer-owned AI API tokens.
The service enables software engineers to launch and oversee containers, database clusters, applications, and artificial intelligence workloads without having to maintain the underlying server infrastructure. The breach disclosure came on 30 October following a warning from cloud security firm Wiz.
Compromised Environment Variables
Following the initial warning, Northflank instructed clients to proceed under the assumption that their environment variables had been decrypted and any confidential data contained inside had been obtained by unauthorised individuals.
Environment variables are routinely relied upon by software teams to maintain delicate connection strings, including database login credentials, third-party authentication tokens, and cloud infrastructure keys. The unauthorised retrieval of these details leaves related systems exposed to broader intrusions, data theft, and unexpected computing charges.
Incident Timeline and Mitigation Steps
Information published by Northflank reveals that Wiz delivered a threat notification at 11:43 UTC on 30 October, with Northflank technicians beginning their investigation sixty seconds later. An initial network policy remediation was deployed shortly thereafter, followed by subsequent patches at 12:50 UTC and 13:05 UTC. In addition, the firm enlisted help from Wiz's dedicated response team.
Northflank has since introduced credential rotation functionality across Redis, MongoDB, MySQL, and PostgreSQL services. It also expanded audit logging capabilities to every tier, including free accounts, so customers can review anomalous activity. Platform users have been instructed to rotate secrets residing within global groups, secret sets, and standalone workloads, whilst disabling external connectivity to add-ons.
Scope of Compromise and AI Key Misuse
Northflank confirmed that several clients reported illicit activity stemming from AI API keys kept on the platform. Subscribers were urged to examine their external AI provider expenditure and operational logs for irregularities and notify both Northflank and their respective AI vendors.
The hosting firm has not revealed the vulnerability exploited to enter its environment, the total count of organisations impacted, or whether credentials apart from AI tokens were taken. Questions submitted to the organisation regarding the extent of the theft remained unanswered at the time of reporting.
Systemic Targeting of AI Infrastructure
The incident forms part of a wider wave of incursions directed at developer platforms that host artificial intelligence tooling. A study published by Wiz, titled Black October for AI Developer Platforms, highlighted at least nine separate intrusions during the month, naming victims including Warp, OpenCode, Railway, Lightning AI, and Northflank.
Within that group, Railway disclosed that decrypted environment variables might have been exposed, OpenCode experienced leaks of AI credentials, and Warp observed compromised telemetry and agent activity logs. Wiz explained that intruders are aggressively searching for machine learning tokens to execute LLMJacking, a technique where attackers hijack third-party accounts to query high-cost models or resell access on secondary markets.
In two instances reviewed by Wiz, hijacked enterprise accounts generated roughly 2.5 million queries to Amazon Bedrock, incurring approximately $100,000 in unapproved costs. While researchers uncovered technical infrastructure used to resell model connectivity, they could not conclusively tie it to the initial breaches, nor has it been proven that all nine incidents stem from the same threat actors.
By the numbers
When PhishPin scanned the internet for domains imitating Northflank, it found 13 lookalike domains registered by third parties, 7 of which are configured to send e-mail, the usual setup for phishing and payment fraud.
Of those 13, one is parked or listed for sale, which is worth monitoring.
Some of these may be defensive registrations by Northflank itself, so we publish counts and patterns rather than individual domains.
See what is impersonating your brand - request a free exposure audit.
Frequently asked questions
What should affected Northflank users do?
Users should assume their stored environment variables were decrypted, rotate all keys and secrets across workloads and add-ons, review their external AI vendor billing records, and restrict public access to database add-ons.
What is LLMJacking?
LLMJacking is an attack method where adversaries use stolen API credentials to query expensive artificial intelligence models on a victim's account, resulting in steep unauthorised bills or enabling the attacker to resell access.
This article is PhishPin's own rewrite of reporting by BleepingComputer, published 31 October 2026. The facts and context are theirs; the wording is ours.
Read the original article on BleepingComputer →Is your brand being impersonated?
PhishPin finds lookalike domains, fake profiles and phishing kits that copy your brand - and gets them taken down. Get a confidential 24-hour audit.
Get my free audit
