Customer impact

Between approximately 4:00 pm UTC on August 3, 2026 and 9:24 pm UTC on August 5, 2026, a subset of customers in the EU region were unable to view large language model prompt audit logs in AI Trust Layer. The background save operation for prompt log records failed during this period, and those records were not stored.

The impact lasted approximately 2 days, 5 hours, and 24 minutes. Running operations were not affected, but the missing prompt log data is unrecoverable. We sincerely apologize for the disruption and the loss of audit trail data during this window.

Root cause

The root cause was a routing configuration mismatch introduced during a service configuration migration. Two internal configuration sources — one controlling service routing and one serving as the service address registry — contained mismatched destination addresses for the affected EU deployment. This caused audit log upload requests to reach an endpoint that returned errors. Because these errors were not retried, the background save operation silently dropped prompt log records instead of storing them.

Detection

An automated alert then fired at 4:14 pm UTC on August 3, 2026 and the initial investigation began. The issue was escalated after a customer reported missing audit logs on August 5, 2026. The gap between detection and resolutions reflects a gap that will be reflected on in the follow up actions.

Response

At 7:14 pm UTC on August 5, 2026, an incident response channel and video call were opened to investigate the customer impact and routing behavior. By 8:30 pm UTC, the team identified that the service routing configuration did not match the service address registry. The plan was to correct the affected addresses and redeploy the configuration.

At 8:34 pm UTC, the impact was confirmed as missing large language model prompt logs for AI Trust Layer in one EU region, with running operations unaffected. The corrected configuration was prepared and deployed. At 9:24 pm UTC, traffic was observed returning to the affected deployment after the update. The incident was marked mitigated at 9:49 pm UTC.

Follow up

1. We are reviewing and updating our automated end-to-end monitoring to ensure that failures in prompt log capture are detected promptly after service updates in all regions.
2. We are conducting a detailed impact analysis to understand the full scope of affected customers and features during the impact window.