OpenAI Codex outage locks out developers for over an hour
At 4 p.m. Friday, Codex stopped working. Two hours earlier, OpenAI had admitted its AI agent leaked user data. Suddenly, every Codex interface failed. Developers everywhere saw 401 Unauthorized errors. For more than an hour, the main coding tool for Silicon Valley was dead. The impact was instant.
OpenAI's status page officially marked the Codex incident as a 'Full outage,' with all impacted services fully recovered and the issue resolved at 11:54 p.m. UTC on September 25, 2026.
Developers who trusted Codex to run their workloads were suddenly cut off. Aiden Bai tried to reconnect again and again. He got the same error each time. Some users who never logged in with API keys were logged out anyway. Forum threads filled with guesses about backend key failures. Nobody had answers. Codex was down. Work stopped.
Social media blew up. Databricks' Yuchen Jin posted a meme about the chaos. OpenAI's research VP, Aidan Clark, said he only heard about the outage from user complaints. At first, he blamed airplane WiFi. The outage hit late afternoon in California. In other time zones, it wrecked the start of the workday. Hours of setup and upgrades were lost.
Reuters reported that OpenAI was still working to understand the full scope of agent activity after a separate data-leak issue involving research and evaluation environments, including user-uploaded images being exposed to third-party services. The company uses anonymized user data in its model-training process, which contributed to the concern surrounding the leak.
After the outage, the community reacted with sarcasm and quick jokes. Some called it a "work-life balance feature" that forced everyone to start the weekend early. Others wanted compensation. One user suggested a banked reset for every 10 minutes of downtime. When Tibo finally announced a full quota reset for paid users, people rushed to use up their quotas before the reset hit.
The outage showed how much the industry now depends on AI agents. Over 90% of Codex's own code is written by Codex. Even the engineers fixing the system needed backup agents. When Codex failed, some developers switched to Claude Code. Most quickly remembered why they left it. The rivalry between OpenAI and Anthropic was clear. Claude had issued its own banked reset earlier that week. Some users joked OpenAI should just let Codex stay down.
One detail stood out. The "Tibo Reset Tracker"-a community tool-monitors when Codex quotas reset. Before this outage, 1,690 developers had already clicked the prayer button, hoping for another free reset. The tracker's chart went flat during Tibo's sleeping hours. Even with AI, human schedules still matter.
This Codex outage shows the risk of automating critical workflows. When the system writes its own code and then crashes, developers are stuck. They wait for a reset or scramble for alternatives. Generous compensation may help for now. But with over thirty incidents on OpenAI's status page this month, patience is running out. AI-driven productivity is tempting. The cost of failure is real.
