TL;DR
Cloud platform Northflank disclosed a breach on 30 October and told customers to assume every secret held in environment variables was stolen, teiss reported on 2 November. Users have since reported misuse of AI API keys. The wider concern is a pattern: security firm Wiz counts at least nine incidents at AI developer platforms during October, which matters to any UK business whose teams hold keys to paid models.
Northflank’s published timeline runs from an 11:43 UTC alert by Wiz, through three patches, to its customer warning at 13:58. The report of unauthorised key use followed late that evening. It has not said how attackers got in, how many customers were hit or whether other credentials were abused, and teiss says its questions went unanswered.
A month of similar incidents
Wiz’s report, titled “Black October for AI Developer Platforms”, also names Railway, OpenCode, Warp and Lightning AI. The details vary. Railway may have exposed decrypted customer variables, OpenCode disclosed exposed keys, and Warp’s breach reached telemetry and agent logs.
The common thread is what the keys are worth. Wiz describes LLMjacking, in which stolen credentials are used to run AI models on someone else’s account, either to resell the access or simply to consume it. In two cases it investigated, compromised enterprise credentials drove about 2.5 million Amazon Bedrock calls and roughly $100,000 of usage.
In our reading, a stolen API key can turn directly into someone else’s bill, so the damage may show up first on an invoice. That is why Northflank’s advice included checking usage and billing records at each AI provider, as well as rotating secrets and revoking old credentials.
What this means for UK teams
The report names no UK customer or regulator, and we have not established whether any UK organisation was affected. The practical lesson applies regardless of location. In our view, teams building with AI have often treated model keys as a developer convenience, stored alongside other configuration, when they now behave more like a corporate payment credential.
Three steps follow from Northflank’s own guidance. Rotate anything held on an affected platform and revoke the old value with its issuer. Review usage and billing with your AI provider. Then ask which other third-party platforms hold your keys, because the Wiz list suggests the exposure is not limited to one vendor.
Looking Forward
Northflank has made audit logs available to every customer, free accounts among them, and enabled credential rotation for its database add-ons. Further disclosure from the company would help, since the attack route remains unknown.
In our view, spending limits and per-project keys are worth adopting now, because they cap the harm from a theft you may not hear about for hours. As more business software comes to depend on metered AI services, the keys to those services are likely to remain an attractive target.