Cloud hosting platform Northflank told developers on 30 October 2026 to rotate their credentials after discovering a security breach that compromised stored environment variables. The Northflank security breach exposed sensitive deployment secrets, enabling attackers to hijack third-party AI API keys and run up unauthorized usage on external model providers.
Key points
- Northflank alerted users that attackers likely decrypted environment variables holding credentials and third-party secrets.
- Multiple developers reported that intruders abused stolen AI API keys to query external machine learning models.
- The compromise follows a wave of October intrusions across cloud developer tools targeting AI credentials, a tactic known as LLMJacking.
- Northflank deployed network patches, enabled automated database credential rotations, and granted free audit log access to all accounts.
How the breach unfolded
Northflank provides developer infrastructure to host applications, microservices, databases, and machine learning workloads without managing bare servers. Because customers run active software pipelines on the platform, its systems store configuration values and access keys needed to connect to external databases and paid APIs.
The company disclosed a security incident after cloud security firm Wiz warned its security team of suspicious activity at 11:43 UTC on 30 October 2026. Northflank engineers investigated the alert one minute later and pushed an emergency network policy patch. The team followed up with two additional patches at 12:50 UTC and 13:05 UTC while bringing in external incident responders from Wiz to assist.
Shortly after patching, Northflank issued an urgent directive to customers. The company instructed teams to operate under the worst-case assumption regarding their stored settings: “Assume all environment variables have been decrypted. Disable/rotate all keys,” Northflank said in a customer advisory.
Unauthorised AI key abuse
Environment variables typically hold sensitive authentication secrets, such as database passwords, cloud tokens, and third-party API credentials. Exposing those plain-text values gives attackers direct pathways into remote services and paid software accounts associated with the compromised projects.
Within hours of the alert, customers began finding unexpected activity on their accounts. “Users have reported unauthorised use of AI API keys supplied to the Northflank platform,” the company said in an updated incident report.
Northflank urged users to review their external AI vendor billing statements and consumption dashboards immediately. Attackers with access to proprietary API keys can submit queries directly to remote model endpoints, draining account credits or billing large fees to the victim before detection.
The rise of LLMJacking
The compromise at Northflank forms part of a broader pattern of cyber attacks targeting development environments that house commercial AI credentials. Security researchers have tracked increased adversary interest in cloud platforms precisely because they store high-privilege tokens for services from OpenAI, Anthropic, and Amazon Web Services.
In a technical report titled Black October for AI Developer Platforms, Wiz documented at least nine security compromises against developer-focused cloud services during the month. The affected platforms included Northflank, Railway, OpenCode, Warp, and Lightning AI.
The incident at Railway involved the possible decryption of customer environment variables, while OpenCode confirmed leaked AI credentials. Warp discovered an intrusion that exposed developer telemetry and AI agent execution logs. Researchers refer to the unauthorised use of stolen enterprise model credentials as LLMJacking.
Attackers use LLMJacking to run expensive machine learning inference workloads at someone else’s expense, or route queries through private proxies to resell model access on dark web forums. In two enterprise compromises analysed by Wiz earlier in the month, attackers made approximately 2.5 million requests against Amazon Bedrock endpoints using compromised access keys, racking up roughly $100,000 in computing fees.
Remediation and unanswered questions
Northflank took several technical steps to contain the damage and help customers assess exposure across their application fleets. The company enabled self-service credential rotation for managed database instances running PostgreSQL, MySQL, MongoDB, and Redis.
Northflank also turned on platform audit logging for every customer tier, including non-paying accounts, to help developers trace when their configuration settings were accessed. Security advisers told users to review global secret groups, workload environment variables, and disable public network access to database add-ons.
The company has not yet confirmed the initial attack vector, nor has it disclosed the total number of customer accounts accessed during the incident. Northflank continues to investigate alongside external security specialists to establish whether other types of stored secrets were extracted.
