ChatGPT and Codex outage
- OpenAI suffered a major global outage that disrupted both ChatGPT and the Codex coding assistant. - OpenAI publicly said it was investigating the disruption affecting users worldwide. - The incident highlighted operational fragility in large AI platforms and renewed focus on reliability engineering for model services (openthemagazine.com).
OpenAI spent much of Monday, April 20, restoring ChatGPT and Codex after users worldwide were suddenly locked out of both services. (status.openai.com) OpenAI’s status page said at 8:13 a.m. ET that affected users were “unable to access ChatGPT, Codex and API Platform,” then posted repeated investigation updates at 8:29 a.m., 9:03 a.m. and 9:48 a.m. ET. The company shifted to monitoring at 10:43 a.m. ET and marked the incident resolved at 11:48 a.m. ET. (status.openai.com) The outage touched more than the chat box. OpenAI listed problems across Conversations, Login, Search, Voice mode, File uploads, GPTs, Image Generation, Deep Research, Agent, Connectors/Apps and the ChatGPT app itself. (status.openai.com) User reports spiked fast on outage trackers as the disruption spread across the United States, the United Kingdom and other markets. One widely cited tally put reports above 5,000 within the first hour, while another said complaints later topped 8,000 at peak. (techresearchonline.com, eweek.com) Codex is OpenAI’s coding agent, and the company had expanded it just four days earlier with an in-app browser, computer-use controls for macOS apps, pull-request review tools and scheduled thread automations. Monday’s disruption hit a product OpenAI has been pushing deeper into everyday developer workflows. (developers.openai.com) The outage also reached OpenAI’s application programming interface, or API, the back-end service other software uses to call OpenAI models. When that layer fails alongside ChatGPT and Codex, the problem extends beyond OpenAI’s own website to businesses that build on the platform. (status.openai.com) This was not an isolated service wobble. OpenAI’s status history shows access issues on April 7, April 10 and April 20, plus a separate April 20 incident for “elevated errors for ChatGPT conversations in Europe.” (status.openai.com, status.openai.com, status.openai.com, status.openai.com) OpenAI said only that it had “applied the mitigation” and was monitoring recovery; it did not publish a cause on Monday. The company has used fuller root-cause writeups after some past outages, including a November 2025 website outage that affected ChatGPT, APIs and Sora. (status.openai.com, status.openai.com) By late Monday morning in the U.S., OpenAI’s public status page showed the affected services back to operational. For users who treat ChatGPT and Codex as daily tools, the day’s main fact was simpler: one outage took out the assistant, the coding agent and the developer pipe behind them at the same time. (status.openai.com, status.openai.com)
