Summary
On June 2, 2026, approximately between 9:46 PM PDT and 12:01 AM PDT the next day, some customers experienced elevated errors and latency across several services, including Codex, ChatGPT, and the Responses API. For affected Responses API traffic, requests took longer than normal to begin generating responses, resulting in increased latency. Codex requests were incorrectly rejected with HTTP 429 responses, and ChatGPT users encountered login, authentication and conversation failures.
Impact
During the incident:
Pay-As-You-Go and Priority Processing Responses API requests experienced elevated latency.
Codex requests received unexpected HTTP 429 responses.
ChatGPT login, authentication and conversation flows experienced partial degradation.
Root Cause
The incident was triggered by a configuration rollout that caused an internal service to repeatedly establish connections to a shared backend dependency. This behavior consumed capacity within shared infrastructure and degraded communication with several downstream services.
Responses API traffic requests to a backend dependency experienced elevated latency before model processing began, resulting in increased latency and failures.
Codex was impacted through a separate dependency that supports request rate limiting. When that dependency became unhealthy, requests were incorrectly treated as rate-limited and returned HTTP 429 responses.
ChatGPT was impacted because degradation within shared infrastructure affected authentication and account-related services, resulting in login, authentication and conversation failures for some users.
Mitigation and Recovery
After identifying the issue, we disabled the affected configuration and redirected traffic away from impacted infrastructure while systems recovered. We also implemented a temporary mitigation for Codex to prevent requests from being incorrectly rejected while the affected dependency recovered.
These actions restored normal service operation. Codex impact was mitigated by approximately 10:35 PM PDT. ChatGPT failures and Responses API latency returned to normal by 12:01 AM.
Prevention and Follow-up Actions
We are implementing several improvements to reduce the likelihood and impact of similar incidents:
Strengthening validation, canary deployments, and rollout controls for infrastructure configuration changes.
Improving safeguards and resilience around shared infrastructure dependencies.
Expanding monitoring and automated detection for infrastructure resource saturation and service degradation.
We apologize for the disruption and are continuing to strengthen the reliability and resilience of our products.
