OpenAI Codex goes down with incorrect API key errors

By Splitfeed EditorialLast updated Sep 26, 2026, 18:55 UTC

On September 25, 2026, OpenAI's Codex service went down and began returning incorrect API key errors, technically classified as 401 authentication errors, to users attempting to access the platform. The disruption lasted approximately 56 minutes. Reports first surfaced on Hacker News, where a user noted the errors and observed that OpenAI's status page initially showed nothing. An incident entry was later added at status.openai.com under the identifier 01M3DCNWMW57HYK8FJ5FBFPA. During the outage, users who authenticated through direct API-key login rather than other methods were able to work around the problem.

The mismatch between the actual outage and the delayed status page update is the detail most likely to draw scrutiny. For developers and teams relying on Codex in production workflows, a 56-minute window of misleading error messages, ones pointing to user-side credential problems rather than a platform failure, can mean wasted debugging time and eroded trust. The core question the incident raises is whether OpenAI's incident detection and communication processes are fast enough to match the scale of its commercial deployments.

No published reactions from the Pro-AI, Anti-AI, or Middle Ground camps have appeared in connection with this specific event. Pro-AI voices would typically treat a sub-hour outage as unremarkable and point to the existence of a workaround as evidence of system resilience. Anti-AI voices would typically use the incident to argue that critical infrastructure built on third-party AI services carries reliability risks that enterprises underestimate. Middle Ground observers would typically focus on the status page lag as a process problem worth fixing, separate from any broader judgment about AI services.

The next thing worth watching is whether OpenAI publishes a post-incident report explaining the root cause of the 401 errors and the delay in status page detection. That report, if it appears on the incident page at status.openai.com, would clarify whether the failure was in authentication infrastructure, monitoring, or somewhere else entirely.