Northflank reports control-plane secret exposure and requires key rotation
Northflank says the first known decryption of control-plane secrets was on October 30, 02:39–03:08 UTC, with no sign of workload-data access. It reports AI-key abuse; rotate all integration and project secrets; use available provider IP/budget limits. Forward-deployed and dedicated control planes are reportedly unaffected; independent forensics continue.
- Disclosed
- Oct 30, 2026
- Updated Nov 6, 2026
- Severity
- Septic Tank Blew
- No CVE: a breach, not a bug
- Splash zone
- Northflank control-plane secrets; the vendor excludes forward-deployed and dedicated control planes
- Northflank · cloud · credentials
- In the wild
- Yes
- Northflank reports multiple users confirming AI-key use; total scope remains unknown.
What happened
The rotation precaution does not prove every secret was stolen. Northflank says evidence focuses on control-plane secrets, with no sign that workloads or their data were accessed; it says forward-deployed and dedicated control planes were unaffected. Confirm your deployment type with support if uncertain. Its belief that the vector was contained within an hour of the first alert is not a guarantee that stolen keys can no longer be abused. Independent forensic review is ongoing, with a root-cause report planned after findings are confirmed; the entry point, attacker identity, full exposure window and victim count remain unpublished.
On November 6, Northflank placed the first known secret decryption between 02:39 and 03:08 UTC on October 30, hours before the 11:43 UTC alert. It reports patching within 60 minutes of notification and notifying customers within two hours. Those are vendor response claims, not confirmation of a complete exposure window or of every credential being revoked. It says workloads remained online throughout; intermittent platform degradation accompanied simultaneous customer redeployments and its own key rotation and hardening.
The fleet patch for CVE-2026-98432 is described separately, not as the cause or a credential-recovery fix. Northflank says it worked with providers to revoke credentials where possible; this does not replace rotating all integration and project secrets yourself. Rotating only database credentials leaves other stored keys in scope.
Crime-scene timeline
Oct 30, 02:39–03:08 UTC
First known secret decryption, according to the later investigation
Northflank disclosed this interval on November 6; it is not a confirmed start or end of the full exposure window.
Oct 30, 11:43 UTC
Wiz alert prompts investigation
Oct 30, 13:40 UTC
Northflank publicly requests immediate rotation
Oct 30, 15:12 UTC
Postgres and MongoDB secret rotation enabled for all users
Oct 30, 19:28 UTC
Broad rotation guidance remains; investigation ongoing
Oct 30, 23:38 UTC
Provider-side revocation and abuse reporting reiterated
Oct 31, 01:04 UTC
Audit logs enabled for every customer, including free users
Oct 31, 13:34 UTC
MySQL and Redis secret rotation enabled for all customers
Oct 31, 22:32 UTC
Minio addon secret rotation enabled for all customers
Oct 31, 23:31 UTC
RabbitMQ rotation enabled for all customers
Nov 2, 11:09 UTC
Control-plane scope and ongoing independent forensics disclosed
Northflank believes the vector was contained within an hour of the first alert and reports no sign of subsequent attacker presence or workload-data access. It excludes forward-deployed and dedicated control planes, reiterates rotation of all integration and project secrets, and says customers have reported AI-key abuse.
Nov 6, 10:09 UTC
Decryption timing, response sequence and provider controls clarified
Northflank reports patching the exploit within 60 minutes of notification and notifying customers within two hours. It says workloads stayed online, attributes intermittent degradation to simultaneous customer redeployments and its own key rotation and hardening, and reports working with providers to revoke credentials where possible. Investigation and independent forensics continue; users still need to rotate secrets, inspect provider logs, and enable available IP restrictions and budget limits.
Who said what
We list each claim and who made it. Primary sources (the vendor, the researcher, the registry) are marked. Where reports disagree, we grade by the worst credible case.