ChatGPT Outage Leaves Users Unable to Access AI Tool
Thursday morning, thousands of people typed a prompt, hit enter, and got nothing back. No spinning dots that turned into an answer. Just errors. Some couldn’t load old chats.
Others couldn’t log in at all. One of the jokes going around online was that everybody would have to use their own brain at work for a few hours.
The complaints started stacking up fast. Downdetector had passed 5,000 reports by 7:53 a.m. Pacific. Ten minutes later, the count had shot past 22,000.
OpenAI took a while to admit something was wrong
When the first wave of reports hit, OpenAI’s status checker still said the company wasn’t aware of any problems. Anyone checking it got a page full of green while their chat window sat frozen.
That changed soon after. The status page updated to say the team was investigating the issue for the listed services, and those services were ChatGPT and Codex.
The incident got filed under the title “Elevated errors across ChatGPT and Codex” and was marked as degraded performance.
Degraded performance sounds mild. The list underneath it wasn’t.
Nearly every corner of ChatGPT went sideways
OpenAI’s own incident record flagged 19 broken components. Fifteen of those sat inside the ChatGPT group alone:
- Conversations
- Login
- ChatGPT Work
- Codex in ChatGPT Desktop
- Compliance API
- Search
- File upload
- Voice mod
- GPTs
- Image Generation
- Deep Research
- Agent
- ChatGPT Atlas
- Sites
- Connectors and Apps
Read that list again, and you get why the day fell apart for so many people. If Conversations and Login both break, there’s no workaround left. You can’t switch to voice. You can’t upload a document instead. You can’t hop into a custom GPT you built. The whole product goes quiet at once.
Some visitors saw stranger behaviour too. The site tried to push an odd file download at people who landed on it, which is never a comforting thing to see from a service you trust with your work.
The problem didn’t stop at ChatGPT
Plenty of apps and websites don’t run their own AI. They plug into OpenAI in the background and put their own logo on top. When OpenAI stumbled, those tools stumbled with it.
Writing assistants, support chat widgets, coding helpers; all of them started throwing errors for reasons their own users couldn’t see.
Developers using Codex got hit from both directions, since the coding side was named in the same incident.
A few people also reported hiccups with rival chatbots like Anthropic’s Claude and X’s Grok around the same window, though those looked smaller and kept working for most visitors.
This was the fourth wobble in five days
Here’s the part that should worry anyone running a business on top of these tools. The Thursday outage wasn’t a rare event. It was the second incident OpenAI logged that same calendar day.
Just after midnight, the company opened an investigation into high error rates in ChatGPT Work Mode, then declared everything recovered six minutes later. The day before, on 2 September, new account creation broke.
On 1 September, the Responses API ran slow, and Free and Go plan users hit errors in their conversations. On 31 August, ChatGPT Work started throwing errors and latency around 11:04 a.m. Eastern and wasn’t declared fully recovered until 8:28 p.m.
Outage trackers put numbers on the pattern. StatusGator counted 61 OpenAI ChatGPT incidents in 90 days, with a median duration of roughly two hours and forty minutes, and 349 OpenAI outages since January 2025. IsDown has logged 192 ChatGPT incidents since October 2025.
That works out to more than one incident every two days across the product line.
How to tell if ChatGPT is really down or if it’s just you
Before you restart your router in frustration, run through this:
- Check status.openai.com: It’s slow to update sometimes, but when it flips to investigating, you have your answer.
- Check Downdetector: User reports spike minutes before official pages catch up, so it’s usually the earlier signal.
- Try a different network: Switch to mobile data or a hotspot. If the site loads there, the problem sits with your connection.
- Turn off your VPN: A blocked or flagged exit node will look exactly like an outage.
- Try another browser or device: This rules out a broken extension, a stale cookie, or a cached login token.
If the site works everywhere except your machine, you’re troubleshooting your own setup, not an OpenAI incident.
What to do while you wait it out
- Stop hammering the send button: Repeated retries during an incident do nothing except add load.
- Copy your prompt before you send it: Long, carefully built prompts vanish when a request fails mid-flight, and rewriting one from memory hurts.
- Keep a second tool ready: Claude, Gemini, and Perplexity all went right on working for most people during this one.
- Save your important outputs somewhere outside the chat window: Chat history is the first thing to disappear when Conversations breaks.
If your team runs customer-facing workflows through the API, write down a manual fallback now, while nothing’s on fire.
Conclusion
ChatGPT crossed 800 million weekly users by late 2025, and OpenAI reports over two billion queries a day flowing through it.
A tool that big has quietly become plumbing for a lot of jobs. Support teams draft replies with it. Marketers outline campaigns in it. Developers ship code through Codex.
Plumbing gets noticed on the day it backs up.
Nothing about this outage was catastrophic. Most incidents like it clear inside a few hours. What it exposed is how many people had no plan B ready when a single status page turned yellow. That’s the fixable part, and it costs nothing to sort out before the next one.