On April 27, 2026, between approximately 11:14 UTC and 12:52 UTC, customers hosted in our Australia region experienced errors and timeouts when using the Intercom REST API. Requests to the API would fail with timeout errors. Fin, the Intercom Inbox, and the Messenger were not affected during this window.
A change was recently shipped with the aim of improving the reliability of our billing system - this had the unintended consequence of triggering in infinite loop for a specific subset of customers, exhausting server resources and causing API failures for unrelated AU customers
No customer data was lost during the incident. Conversations, messages, and other records remained intact throughout. The issue was limited to the ability to serve API requests in real time — once the issue was resolved, all functionality returned to normal with no lasting effects.
We understand that the Intercom API is core infrastructure for how many of you operate your businesses, and that any period of unavailability directly impacts your customers and your teams. We sincerely apologize for the disruption this caused.
11:14 — API error rates begin climbing in the Australia region
11:32 — Automated monitoring detects the AU API is unhealthy. Incident declared and engineering paged
11:35 — Responders confirm customer impact
11:37 — Severity escalated to Major. Status page posted: "Errors on the REST API in Australia"
11:42 — Crash source identified in AU logs, linked to the billing cache fallback path
11:43 — Rollback of recent deployments initiated as a precaution
12:26 — Root cause identified: a circular dependency in the billing cache read path triggered when the cache is cold
12:48 — Mitigation applied: restoring the affected customer's cache entry confirmed to stop the crash loop
12:50 — Errors confirmed stopped across the AU API fleet
12:52 — Status page marked resolved: "Confirmed that the fix resolved the issue, service is operating normally"
13:29 — Permanent fix merged
The root cause has been fixed. The problematic code has been replaced, with regression tests added to prevent further occurrences
We are auditing all related code paths within the billing system to ensure no other operations can trigger a similar circular dependency.
We are adding structural safeguards that make this class of recursion impossible regardless of future changes to the system.
