Summary
On July 21, 2026, some users experienced errors when signing in to ChatGPT, the API Platform, and Codex during two periods: approximately 11:18–11:40 a.m. PDT and 3:00–3:10 p.m. PDT. Unintended infrastructure maintenance and subsequent database restarts temporarily disrupted the identity services supporting new login attempts. Services recovered after each restart was completed.
Impact
During these periods:
Some users were temporarily unable to log in to ChatGPT, the API Platform, or Codex.
Users who were already signed in could continue using ChatGPT.
Core model inference and API completions were not directly affected.
Root Cause
Unintended maintenance by an infrastructure provider affected a database supporting our identity services and reduced its backup capacity. The provider subsequently restarted the database on two separate occasions, temporarily disrupting database connections and causing errors for new login attempts across ChatGPT, Codex, and the API Platform. The second disruption was caused by an automated restart.
Resolution
Service recovered after each database restart completed and authentication services re-established their connections. Recovery from the first disruption began at approximately 11:26 a.m. PDT and was complete by 11:40 a.m. PDT. The second disruption was resolved by approximately 3:10 p.m. PDT. Following recovery, engineers coordinated with the infrastructure provider to prevent further unintended maintenance and disable the faulty restart automation.
Prevention and Improvements
We have implemented or are working on several improvements to reduce the likelihood and impact of similar incidents:
Ensure users can sign in during database outages. Existing sessions were unaffected. We reduced unnecessary database dependencies to help keep new login attempts working during future database disruptions.
Strengthen maintenance safeguards. Improve coordination with our infrastructure provider through verified maintenance exclusions, direct escalation paths, and additional checks before disruptive actions.
Improve backup monitoring. Strengthen alerts for degraded backup capacity and improve recovery checks for database connectivity.
Expand resilience testing. Test database restarts and recovery across login, sign-up, organization management, and dashboards.
Prevent unnecessary restarts. Work with our infrastructure provider to improve restart automation and reduce avoidable disruptions.
We apologize for the disruption and are continuing to strengthen the reliability of the systems that support ChatGPT, the API Platform, and Codex.
