Northflank tells customers to rotate every key after AI API key breach

Northflank disclosed a security incident on its platform Oct. 30, 2026, after Wiz sent a threat alert at 11:43 UTC. <sep>

The company told all customers to rotate all credentials and secrets kept on the platform. Customers told Northflank that attackers had used their AI API keys without permission. Northflank said, “Users have reported unauthorised use of AI API keys supplied to the Northflank platform.” <sep>

Northflank advised customers to treat all environment variables as decrypted and the secrets within them as stolen. The company has not explained how the attacker reached the secrets, whether anyone read database data, or how many accounts experienced impact. Northflank said, “no additional confirmed findings to share at this time,” and said its investigation remained ongoing as of 19:28 UTC on Oct. 30, 2026. <sep>

Northflank’s engineers looked at the Wiz alert at 11:44 UTC on Oct. 30, 2026, one minute after it arrived. Northflank put out an initial network policy patch at 12:02 UTC, with additional patching at 12:50 UTC and 13:05 UTC, and requested help from Wiz’s incident response team at 13:00 UTC. It first told customers to rotate environment variables immediately at 13:40 UTC. At 13:58 UTC, Northflank told all users, “Assume all environment variables have been decrypted.

Disable/rotate all keys.” <sep>

Northflank activated credential rotation for PostgreSQL, MongoDB, MySQL, and Redis on Oct. 30, 2026. As of its Oct. 31, 2026 update, the company enabled credential rotation for MySQL and Redis for all customers. Northflank enabled audit logs for all customers, including those on free accounts, at 01:04 UTC on Oct. 31, 2026. The company said customers should rotate secrets in workloads, secret groups, and global secrets, and should turn off public access to add-ons where that option is on. <sep>

Northflank told customers to check AI provider usage records and billing logs for unexpected activity and to report any suspicious use to Northflank and their AI providers. Northflank said billing logs show when customers rotated keys before anyone reported misuse, though someone may have used the keys earlier. <sep>

Major outages hit seven components — networking, certificates, services, add-ons, jobs, builds, and logs and metrics — even though the app, API, marketing site, and documentation stayed online. Northflank’s status page showed major outages for those components on the morning of Oct. 31, 2026. Its 90-day record shows 99.65 percent uptime across the platform. <sep>

Wiz’s Black October report covers at least nine incidents from October 2026 against AI development platforms, including Northflank, Railway, OpenCode, Warp, and Lightning AI. Wiz has not said all nine incidents came from the same attackers, and Wiz said, “there is no single vulnerability or dependency linking these attacks.” At 23:38 on Oct. 30, 2026, Northflank said AI key abuse reports may match the wider attack pattern, pointing to IP addresses listed in Wiz’s Black October report. <sep>

Wiz’s report lists Oct. 6, 2026 for Fireworks AI, Oct. 20, 2026 for Railway, Oct. 23, 2026 for Warp, and Oct. 30, 2026 for Northflank. Fireworks AI, Zeabur, and Lightning AI say attackers got in using stolen credentials. Railway says its attacker got in through privilege escalation in build infrastructure. Warp points to a third-party vendor as how its attacker entered on Oct. 23, 2026, and OpenCode points to a legacy system from Oct. 16-17, 2026; Fireworks AI, OpenCode, and Warp reported signs of an agent swarm, with attackers using AI agents to move faster. <sep>

Wiz said attackers in some incidents went after environment variables that hold AI service credentials. LLMJacking describes using stolen credentials to run AI models on another organization’s API keys, leading to large charges or resale of the access. In two cases studied by Wiz, compromised enterprise credentials generated about 2.5 million Amazon Bedrock API calls valued at about $100,000 in AI usage, and Wiz traced two stolen AWS Bedrock keys to roughly 800,000 calls in 30 hours and 1.7 million calls in 13 hours, mostly for Claude models, and estimates those two cases likely generated more than $100,000 in AI usage. Wiz says stolen keys showed up in resale pools offering low-cost access to AI models, testing began within 30 minutes of theft in the cases it studied, and heavy use began four and seven days later; Wiz shared IP addresses associated with tested or used keys but says it has no evidence linking them directly to the platform breaches. <sep>

Northflank keeps each service’s environment variables encrypted in a shared database, decrypting them in memory when a workload starts or is rebuilt. If an attacker controls the decryption layer, that person can see any secret that passes through it, not just one customer’s. Encryption at rest does not prevent exposure, since a decryption component has to run to deploy an app. Railway’s preliminary report says decryption exposed its encrypted variables despite the encryption. <sep>

Northflank’s Bring Your Own Cloud page says workloads, databases, volumes, and secrets run inside the customer’s virtual private cloud, while the status page does not separate Bring Your Own Cloud customers from others. Northflank has not said whether the incident affects Bring Your Own Cloud customers. <sep>

Railway and Vercel also experienced security incidents in 2026 that concluded with advice to rotate credentials. StartupHub.ai says Northflank customers may end up covering the cost of AI usage burned by someone else. <sep>

Qovery says to start with AI and payment key rotation, move to database and cloud credentials next, cover all remaining secrets, and review provider usage logs from before the incident. Qovery says to create a replacement key, deploy it, confirm the app works, and then revoke the old key. Qovery recommends limiting new AI keys to the models and regions actually used and setting a quota or spending limit. <sep>

Northflank’s guidance remains to change every credential and secret kept on the platform, assume someone decrypted environment variables, and disable or rotate all keys. Northflank instructs customers to review AI provider usage records and billing logs for unexpected activity and to notify Northflank and their AI providers of any suspicious use. <sep>