On the evening of August 19, 2026, ChatGPT went down for users everywhere trying to log in, sign up, or pull up past conversations. The disruption started around 8 PM ET and didn't stop at the chatbot — it reached into other OpenAI products too, including the Codex coding platform and as many as a dozen API endpoints.
The Root Cause
Per OpenAI's status page, engineers traced the trouble to a failure retrieving authentication verification keys — the process that confirms who's logging in. Without that working, new signups and logins across the affected services just stopped.
What Users Ran Into
People trying to open ChatGPT hit loading animations that spun forever, "too many concurrent requests" errors, and failed message sends. Downdetector reports spiked around 5 PM PT, with trouble showing up on both sides of the Atlantic — users in the US and Europe both flagged issues.
What Still Worked
The outage wasn't total. Testing by 9to5Mac found ChatGPT's Mac desktop app kept working while the web version stumbled, sometimes failing to load conversation history or account projects, or to generate responses at all. Forbes reported that people who were already logged in could generally keep using the core ChatGPT service — it was new logins and signups that broke.
OpenAI's Response Timeline
OpenAI confirmed the issue on its status page at 8:03 PM ET, saying it was working on a fix. The company marked the incident "identified" by 8:15 PM ET — about 14 minutes after it began — though BleepingComputer noted problems were still turning up in its own testing at that point.
Authentication failures like this one point to a familiar weak spot in large cloud platforms: even when the core infrastructure is fine, one broken login system can cut off access for everyone.