Cloud development platform Northflank is strongly advising customers to rotate their credentials after a security breach that may have exposed their stored secrets, including API keys for accessing AI services.
According to its status page, Northflank received a security alert from cloud security firm Wiz on Oct. 30 and immediately began investigating the incident.
Northflank deployed several network policy patches on the same day and warned customers: “Assume all environment variables have been decrypted. Disable/rotate all keys.”
The company subsequently reported that multiple customers had confirmed unauthorized use of their AI API keys.
Customers were instructed to revoke existing credentials, generate replacements, review AI service usage and billing records, and disable public access to add-ons if enabled.
Northflank has also enabled secret rotation for various database and messaging add-ons offered through its platform and made audit logs available to all customers.
AI key theft poses wider financial risks
The recent Northflank breach comes amid a broader wave of attacks targeting AI-based development platforms.
In a report published on Oct. 30, Wiz identified at least nine platforms that had disclosed security incidents in October, including Northflank, Railway, Warp and Lightning AI.
Researchers at Wiz observed that attackers appeared particularly interested in stealing environment variables containing credentials for AI services.
They also documented two distinct cases in which compromised credentials were used to make approximately 800,000 and 1.7 million Amazon Bedrock model calls, respectively.
Wiz estimated that the attackers consumed over $100,000 worth of AI services across the two cases, based on standard service prices. Notably, neither case was specifically attributed to the Northflank breach.
“The use of the keys by multiple actors suggests a robust market for this type of credential,” the researchers wrote.
The Wiz report highlights the financial risks of LLMjacking, where attackers exploit stolen credentials to access AI services without authorization. Wiz’s findings also suggest that restricting API key permissions could limit unauthorized AI model usage, although these restrictions may not prevent it entirely.
