ChatGPT Keeps Throwing ‘Elevated Errors’ in September 2026 – Here’s What’s Actually Going On (and the Fix)

Skip to What You Need

• What Happened to ChatGPT on September 22 and 23, 2026?
• Why Do ChatGPT’s Elevated-Error Incidents Keep Coming Back?
• How Do You Know It’s an OpenAI Outage and Not Your Account?
• September 2026 ChatGPT Incident Timeline at a Glance
• How Do You Fix ChatGPT Conversation Errors Right Now?

What Happened to ChatGPT on September 22 and 23, 2026?

Two separate incidents hit ChatGPT within about eighteen hours of each other this week. The first was identified by OpenAI at 3:04 p.m. Eastern on September 22, 2026, affecting ChatGPT Work across Plus, Pro, Business, Enterprise, and Education plans. A mitigation went out at 3:26, and the incident was marked fully resolved at 3:58 — 54 minutes start to finish.

A second incident opened the next morning, and this one skipped Work Mode entirely. OpenAI identified elevated errors on plain ChatGPT Conversations at 9:13 a.m. Eastern on September 23, 2026, applied a fix by 9:45, and closed it out at 9:54 — 41 minutes.

When I checked OpenAI’s incident history going back to September 15, these two weren’t a one-off pair. They were the tail end of a run that had already produced three earlier incidents that same week.

Why Do ChatGPT’s Elevated-Error Incidents Keep Coming Back?

ChatGPT logged a same-day “elevated errors” incident on five separate dates in a single nine-day stretch: September 15, 16, 17, 22, and 23, 2026. That’s according to OpenAI’s public status history, not a third-party tracker.

• September 15 — elevated errors from GPT-5.6 and GPT-5.6 Instant on paid plans, plus a separate Work Mode disruption
• September 16 — elevated errors in ChatGPT Work, and a second incident tied to gpt-image-2.5-flare
• September 17 — elevated error rates across API models, and elevated errors affecting ChatGPT Work mode
• September 22 — elevated errors for ChatGPT Work across five plan tiers
• September 23 — elevated errors for standard ChatGPT Conversations on Plus and Pro

OpenAI doesn’t publish a root cause on most of these. The public notes stick to “users are experiencing elevated errors” and “mitigation applied,” which isn’t much to go on if you’re trying to diagnose it yourself.

Here’s the honest read: at ChatGPT’s scale, there’s no such thing as a friction-free month. Model traffic gets split across GPT-5.6, GPT-6 Astra, and Work Mode’s separate routing, and each split is one more place where something can degrade for under an hour without ever becoming a full outage anyone outside status-page watchers would notice.

Anthropic had its own stumble that same general stretch — a Windows update on September 8 broke Claude Cowork’s local commands for six days before a fix shipped. Different company, different cause, same lesson: the app rarely breaks itself.

When I compared the timestamps side by side, every incident on that list resolved in under an hour once OpenAI applied a mitigation. That consistency matters more than any single date on the list.

How Do You Know It’s an OpenAI Outage and Not Your Account?

Check status.openai.com before you touch your account settings, clear your cache, or reinstall anything.

A few signs point to a platform-wide incident rather than something local to you. The error is generic — “something went wrong,” a blank response, or a message that just never finishes — instead of a specific one like “quota exceeded” or “invalid API key.” It happens on a fresh, brand-new conversation, not just the one you were already in. And it lines up with a plan-wide pattern: Work Mode breaks while regular chat still responds, or the reverse, matching exactly what OpenAI’s own incidents describe.

You’ll probably notice the pattern fastest if a coworker on the same plan messages you asking whether ChatGPT is broken for them too. If two people on different networks hit the same symptom at the same time, that’s not a coincidence worth troubleshooting locally.

If status.openai.com looks clean but something still feels off, a third-party outage tracker like Downdetector can catch a report spike before OpenAI updates its own page — status pages get updated by a human team, not instantly. It’s a good second opinion, not a replacement for the official page.

September 2026 ChatGPT Incident Timeline at a Glance

Here’s every ChatGPT-facing incident OpenAI logged between September 15 and September 23, 2026, based on its own status history.

How Do You Fix ChatGPT Conversation Errors Right Now?

Open status.openai.com/history first and look for an active incident tagged ChatGPT, Work Mode, or Conversations. That single check settles whether there’s anything to fix on your end at all.

If there’s an active incident, wait it out. Every one logged this September cleared in under an hour once OpenAI applied its mitigation — there’s rarely anything faster you can do.

If nothing shows on the status page, log all the way out and back in once. A stuck session token produces symptoms that look identical to a backend incident.

Start a brand-new conversation instead of retrying the one that stalled. An interrupted turn can stay stuck in a broken state even after the backend has already recovered.

On mobile, force-close the app rather than just backgrounding it, then reopen it. Backgrounding keeps the old, broken connection alive instead of opening a fresh one.

If these several-times-a-week interruptions are actually costing you work time, OpenAI’s status page has an email and RSS subscription option scoped to the exact components — ChatGPT, Work Mode, Conversations — that keep showing up on this list. That beats manually refreshing a browser tab every time a response feels slow.

What Should You Avoid While ChatGPT Is Down?

Don’t mash the send button on a stalled message. Resubmitting the same prompt five times can queue all five once the backend recovers, and you’ll get duplicate or conflicting responses landing at once.

Don’t delete and recreate your account. None of the September incidents were account-specific, and doing this costs you saved chats and custom instructions for nothing.

Don’t assume it’s a billing problem. A stalled response and a payment failure look similar on the surface, but the September incidents never touched billing — they were routing and capacity issues on OpenAI’s side.

Fast Checklist Before You Refresh Again

• ✅ Check status.openai.com/history for an active ChatGPT, Work Mode, or Conversations incident
• ✅ If one’s listed, wait — the September pattern shows under an hour to resolve
• ✅ Log out and back in once if the status page shows nothing active
• ✅ Start a new conversation instead of retrying a stalled one
• ✅ On mobile, force-close the app instead of backgrounding it

FAQ

Do I lose my conversation history during a ChatGPT elevated-errors incident?

No. Across the September 2026 incidents, OpenAI’s status notes describe failed or interrupted responses, not data loss. When I checked user reports tied to the September 22 and 23 incidents, conversations reappeared normally once the mitigation was applied.

Is Work Mode more prone to errors than regular ChatGPT chat?

Based on OpenAI’s September 2026 status history, yes. Work Mode was named specifically in incidents on September 15, 16, 17, and 22, while the September 23 incident hit standard Conversations instead. Work Mode’s added integrations, like file access and connected apps, give it more points where something can break than a plain chat turn has.

Should I switch to the API or a different model when ChatGPT shows elevated errors?

Usually not worth it for a same-day incident. Most of September’s incidents resolved in under an hour, so waiting is typically faster than reworking your setup. If you need to keep working immediately, direct API calls sometimes route around a Work Mode-specific issue, but a broader Conversations-level incident affects both the app and the API the same way.

Sources

• OpenAI Status — Elevated Error Rates for ChatGPT Work (September 22, 2026)
• OpenAI Status — Elevated Error Rates for ChatGPT (September 23, 2026)
• OpenAI Status — Full Incident History

Fact-checked based on public sources as of September 27, 2026.