On July 19th, 2026, between approximately 7:08 AM and 8:05 AM PDT, some ChatGPT users were unable to load or continue conversations, and some Codex requests failed. Cloud infrastructure maintenance reduced capacity for an internal service in one region, leading to elevated errors. Engineers redirected traffic away from the affected region, and service was fully restored by approximately 8:05 AM PDT.
During this period:
Some ChatGPT users could not load or continue conversations.
Voice, Work Mode, Connectors, files and library, and some Codex requests also experienced errors.
Recovery began around 7:46 AM PDT, and service was fully restored by approximately 8:05 AM PDT.
During cloud infrastructure maintenance, a regional database replica supporting an internal identity service became unavailable. The remaining regional database capacity was not sufficient to handle the full workload, causing database requests to slow or fail.
These failures reduced cache effectiveness and increased load across dependent services. Existing failover behavior did not automatically redirect enough traffic away from the affected region, so protective controls began rejecting requests to prevent further overload.
Engineers began manually redirecting traffic from the affected region to healthy capacity in other regions at approximately 7:46 AM PDT. This reduced load and allowed affected services to recover. The cloud provider paused further maintenance while engineers validated recovery and planned capacity rebalancing. Recovery began after 7:46 AM PDT, and service was fully restored by approximately 8:05 AM PDT.
We are working on several improvements to reduce the likelihood of similar incidents:
Strengthening regional capacity planning and redundancy, including explicit validation that services can withstand the loss of a database replica.
Improving automatic failover and cross-region traffic rerouting when regional infrastructure becomes unhealthy.
Hardening retry and circuit-breaker controls, supported by expanded failure testing and continuous topology verification.
We apologize for the disruption and are continuing to strengthen the resilience of the systems that support ChatGPT and Codex.
