Cloud platform Northflank has revealed that attackers were decrypting its customers' secrets hours before anyone raised the alarm, in a new update on the security incident it disclosed on 30 October. The company has also patched a Linux kernel flaw, CVE-2026-98432, across its fleet, although it has not said whether that flaw was how the attackers got in.

Northflank provides a managed platform for building, deploying and running applications, containers, databases and AI workloads. Its customers store environment variables on the platform, which often hold the keys and passwords applications need: database credentials, cloud access keys, authentication tokens and AI provider API keys.

Secrets taken hours before the alarm

The timeline published by Northflank began with a threat alert from cloud security company Wiz at 11:43 UTC on 30 October. Engineers reviewed it within a minute and applied patches, and by 13:58 UTC the company was telling customers to assume every stored environment variable had been decrypted and to rotate all of their secrets. Later that day, users reported that AI API keys supplied to the platform were being used without authorisation.

A further update on 6 November added a troubling detail: the first known decryption of secrets happened between 02:39 and 03:08 UTC on 30 October, eight to nine hours before the Wiz alert arrived. Northflank says the exploit was patched within 60 minutes of the alert and customers were notified within two hours. The company has not published a root cause, and independent forensic auditors who began reviewing the attack over the following weekend are still working. A root cause analysis is promised once findings are confirmed.

What was, and was not, affected

According to Northflank, the attack was contained within an hour of the first alert, and there is no sign the attacker remained in its environment afterwards. Running workloads were unaffected, and the company says it has seen no sign that workloads or workload data were accessed. The evidence, it says, points to an attack focused on secrets held in its control plane, the shared management layer of the platform. Customers on forward-deployed or dedicated control planes were unaffected.

Separately, Northflank has patched CVE-2026-98432 across its fleet, including bring-your-own-cloud clusters where it has an active integration. The company describes the patch separately from the breach and has not linked the flaw to the intrusion. It has also revoked Northflank access tokens and sessions, version-control integrations and cross-account cloud roles where it could act on customers' behalf.

Northflank has not said how many customers were affected, how the attackers gained access, or whether credentials other than the reported AI API keys were stolen or abused.

What customers must do now

Northflank is telling every customer to rotate all secrets now, both integration secrets and project secrets, even if they have seen no suspicious activity. It recommends starting with AI provider keys, cloud provider credentials, payment provider keys, private keys and database credentials, then reviewing provider logs for unusual activity. Customers who spot unexpected AI usage or spending should contact their AI providers directly, such as OpenAI and Anthropic.

The incident is part of a wider wave of attacks on AI developer platforms. In a report titled Black October for AI Developer Platforms, Wiz counted at least nine such platforms that disclosed incidents during October, including Fireworks AI, Railway, OpenCode, Warp and Lightning AI. Attackers in these incidents targeted environment variables holding AI service credentials, which can be abused through a practice known as LLMjacking: using stolen credentials to run AI models on someone else's account, running up large charges or reselling the access. In two cases Wiz investigated, compromised enterprise credentials were used to make about 2.5 million Amazon Bedrock API calls, generating roughly 100,000 dollars in AI usage.

For Northflank customers, the lesson is blunt: assume the keys are compromised, rotate them all, and check the bills.