Summary
On July 14th, 2026, between approximately 4:39 PM and 5:19 PM PDT, some ChatGPT users experienced elevated errors and timeouts when accessing ChatGPT on the web. A web traffic-routing configuration change unintentionally concentrated traffic on a smaller subset of available capacity, causing overload and requests to fail or time out. We reverted the configuration, and service recovered by approximately 5:19 PM PDT.
Impact
During this period, some ChatGPT users experienced errors or timeouts when accessing ChatGPT on the web.
ChatGPT availability on the web temporarily dropped.
Affected users were unable to load ChatGPT, start new sessions, or sign in.
The disruption lasted about 40 minutes before service was restored.
Root Cause
The issue was caused by an internal web traffic-routing configuration change that sent a much larger share of traffic than intended to a smaller subset of available capacity. The resulting concentration of traffic overloaded those resources, causing requests to fail or time out.
Existing safeguards did not prevent the routing change from concentrating more traffic on those resources than they could support.
Mitigation and Recovery
Engineers identified the unexpected routing change and reverted traffic to the standard deployment at 5:18 PM PDT. As traffic returned to its normal destination, service recovered and error and login metrics returned to normal by approximately 5:19 PM PDT.
Prevention and Improvements
We are implementing several improvements to reduce the likelihood of similar incidents:
Strengthening monitoring and progressive rollout controls so unexpected traffic shifts can be detected and stopped earlier.
Adding safeguards that limit how much production traffic can be shifted at once during staged rollouts.
Adding confirmation prompts and impact previews for routing changes that could significantly affect production traffic.
We apologize for the disruption and are continuing to strengthen the safeguards around production traffic changes.
