Microsoft Copilot Is Down With a ‘Trouble Responding’ Error — What Happened on September 17, 2026 and How to Fix It

Copilot didn’t go down quietly. Users on Windows, the Microsoft 365 apps, and the web version all hit the same wall within minutes of each other on September 17, 2026.

What Actually Happened to Copilot on September 17?

Microsoft Copilot started degrading at roughly 13:50 UTC on September 17, 2026, when requests that normally take a few seconds either timed out or returned an error instead of a real answer. According to Microsoft’s own status reporting, the issue was tracked back to a network infrastructure problem, not a content or safety filter change. By the time Downdetector reports crossed 3,000, the pattern was already clear: this wasn’t one region or one app, it was the shared backend that Word, Excel, Teams, and the standalone Copilot app all lean on.

Microsoft said Copilot was back to full operational readiness by 5:21 PM ET that day. That’s roughly a three-and-a-half-hour window between the first reports and the all-clear, though some users reported spotty responses for another hour or so after the official fix landed — infrastructure fixes don’t always propagate to every session at once.

When I checked the error text reported across multiple outlets, it was consistently the same generic line: “I’m sorry, I’m having trouble responding to requests right now.” That phrasing matters. It’s Copilot’s catch-all for a backend failure, not a message tied to your account, your prompt, or your subscription tier.

Microsoft hasn’t published a full root-cause writeup as of this article, only that the issue traced to network infrastructure and that engineers were “provisioning additional capacity” to handle the load once traffic recovered. That second detail is worth sitting with. A pure networking failure usually resolves the instant the bad route is pulled. Needing extra capacity on top of that suggests the backlog of retried requests briefly overwhelmed whatever came back online first — which lines up with the reports of spotty service continuing for close to an hour after the official all-clear.

Is Copilot Down Right Now, or Is It Just You?

A stuck browser tab looks exactly like a server-side outage from where you’re sitting. Here’s the fast way to tell them apart.

Status.cloud.microsoft is the fastest confirmation — it’s Microsoft’s own service health dashboard and it updates faster than most third-party trackers during an active incident.

How Do You Fix It on Your End First?

Most of what looks like “Copilot is still down” after Microsoft posts an all-clear is actually a stuck session on your device. Work through these in order before assuming the outage is ongoing.

1. Hard refresh the tab or app. A cached failed connection doesn’t always clear with a normal reload. Use Ctrl+Shift+R on the web version, or fully close and reopen the desktop app.
2. Sign out and back in. Copilot sessions can hold onto a broken token from the outage window even after the backend recovers.
3. Check status.cloud.microsoft directly. If it shows green for Copilot but you’re still failing, the problem has moved to your device or network.
4. Try a different network. Corporate VPNs and some firewalls cache DNS failures during an outage and need a few extra minutes to pick up the recovered route.
5. Clear the app’s local cache. On Windows, this is usually enough on its own — most post-outage stragglers clear up here.

This part trips a lot of people up: they see the same error an hour after Microsoft’s all-clear and assume the fix didn’t work. Usually it did. Their session just didn’t refresh.

What Happens to Work That Got Cut Off Mid-Response?

If Copilot was mid-draft when the outage hit — a long email, a spreadsheet formula explanation, a Teams meeting summary — that output is usually gone. Copilot doesn’t checkpoint partial generations the way some coding tools do.

A few things worth doing right after an outage like this:

• Check Word and Excel’s built-in AutoRecover before you assume a half-written draft is lost — Copilot’s failure doesn’t always take the document’s own autosave with it.
• Re-run the exact same prompt once status.cloud.microsoft shows green. Outputs aren’t identical between runs, so if the first result mattered, save it as soon as it reappears.
• For Teams meeting recaps specifically, the recording and transcript are stored separately from the AI summary — you can regenerate the recap after the fact without re-attending the meeting.

Honestly, that’s usually enough to recover 90% of what felt lost in the moment.

How Does This Compare to the September 3 Outage?

This wasn’t the month’s first AI outage. ChatGPT, Claude, and Grok all went down together on September 3, 2026, after an Azure East US failure — and that one ran about 90 minutes with Gemini staying untouched because it runs on Google Cloud instead of Azure.

The September 17 Copilot incident looks similar on the surface — infrastructure-level, not a specific feature bug — but it hit a different layer of Microsoft’s stack and ran roughly twice as long. When I compared the two timelines, the pattern that stands out is how concentrated AI tooling has become on a small number of cloud backbones. One Azure region having a bad day in September 2026 was enough to take out three unrelated AI products at once.

Should You Keep a Backup AI Tool for Days Like This?

If Copilot is load-bearing for your actual job — drafting client emails, summarizing meetings you couldn’t attend, cleaning up spreadsheets on a deadline — a single point of failure is a real risk, not a hypothetical one. Two outages in two weeks is enough of a pattern to plan around.

You don’t need a paid backup. A free-tier account on a tool running on different infrastructure (Gemini on Google Cloud rather than Azure, for instance) covers the gap for anything text-based. Keep the login saved somewhere other than a password manager that also lives inside the same ecosystem you’re trying to work around.

Most people skip this and regret it exactly once — usually on the day it actually matters.

There’s a second, quieter risk here too: outages cluster around infrastructure providers, not AI vendors. Anthropic’s Claude and OpenAI’s ChatGPT both went down on September 3 for the same Azure-related reason, even though they’re built by two unrelated companies. Picking a “backup” that happens to run on the same cloud backbone as your main tool doesn’t actually protect you — check where a tool runs before assuming it’s a real alternative and not just a different logo on the same outage.

FAQ

Why does Copilot still show an error after Microsoft says it’s fixed?

Your session likely cached the failed connection from during the outage. Sign out, clear the app cache, and sign back in — that resolves it in most reported cases within a few minutes.

Did the September 17 outage affect Copilot in Word and Excel differently than the standalone app?

No — reports across Word, Excel, Teams, and the standalone Copilot app described the same generic “trouble responding” error, which points to a shared backend issue rather than a per-app bug.

Is there a way to get notified the next time Copilot goes down?

Status.cloud.microsoft doesn’t have a public push-alert feature built in, but third-party trackers like Downdetector let you set email or browser alerts for a specific service, which is faster than waiting to notice it yourself.

Sources

• Android Authority — Microsoft Copilot outage, September 2026
• Windows Report — “Trouble responding to requests right now” error
• Microsoft 365 Service Health Status