OpenAI Outage Hits ChatGPT and Codex, Users Logged Out With 401 Errors

OpenAI's ChatGPT and its Codex coding tool failed for users on Thursday, with sessions abruptly ending and requests rejected by an error reading "Unexpected status 401." The fault touched both the consumer chatbot and the developer product simultaneously, an early signal that the break sat in the account layer rather than inside one application. OpenAI had not published a public explanation for the cause as of reporting.

A 401 response is an authentication failure: the server refused the request because it would not accept the credentials attached to it. For users, that looked like being bounced to a login screen or having a valid session treated as expired without any change on their end. Because two separate products broke at the same time, the likely location is the shared sign-in path or the system that issues and validates access tokens, not a defect confined to one interface.

The Codex side carries the sharper commercial edge. The product has become central to OpenAI's pitch to software teams, sold as a paid tool that plugs into editors and command lines and leans on a continuous connection to OpenAI's servers. When authentication fails, that dependency turns into a hard stop: engineers routing routine work through Codex lose the ability to run tests, merge changes, or ship. ChatGPT, with hundreds of millions of weekly users, produces a far larger volume of visible complaints from even a short disruption, but the developer workflow is where the interruption converts directly into stalled work.

This is not an isolated pattern for the sector. Anthropic's Claude and Google's Gemini have each run into login and capacity failures over the past year, and companies that now push critical workflows through hosted models have begun treating provider downtime as an operational risk rather than a rare annoyance. That shift has driven interest in multi-model configurations, local fallbacks, and status-page monitoring. The switching costs stay high, though, once a team has wired its pipeline around one vendor's API and tooling.

What remains unresolved is the root cause and the duration. Authentication incidents can originate anywhere from a bad configuration push to an overloaded identity service, and OpenAI's status updates will decide whether users file this as a brief blip or the start of a pattern. For developers whose Codex sessions were cut off, the immediate questions are whether unsaved work was lost and whether OpenAI will say more than that its systems are green again.