Situation report: Microsoft 365 is down hard, and the fixes keep failing.
Microsoft is struggling to recover from a long outage that has knocked out Office, Outlook and Copilot, according to The Information. The disruption started Thursday, November 12. By Friday, many users still couldn’t get back in. Other coverage of the incident says Microsoft’s first repair attempts didn’t work as planned.
This is an outage story, but it’s also a story about AI. When the productivity suite goes down, the AI assistant built into it goes down too.
🎯 Threat Assessment
Here’s what we know from the reporting so far:
- Wide scope. Microsoft’s service advisory lists Exchange Online, Teams, SharePoint Online, Microsoft 365 Copilot and Defender XDR as affected.
- Copilot can’t do its job. Copilot Chat may fail to load or answer, and it may lose access to your email, calendar and files. An assistant without your data is close to useless.
- Huge volume. Outlook complaints on Downdetector peaked at about 130,000. New reports have since dropped off for some products.
- Not an attack. Microsoft executive Charles Lamanna said a cyberattack didn’t cause the outage.
- Repairs stalled. Microsoft admitted its remedial changes didn’t work as expected and said it’s trying other approaches. Some regions recovered partially, but the recovery didn’t hold.
- Third-party AI hit too. Reuters reported that the outage also reached connectors that OpenAI’s and Anthropic’s AI services use to pull data from Microsoft accounts.
⚠️ Why This Matters for AI
Microsoft has spent the last few years putting Copilot at the center of its enterprise pitch. Its message to customers was simple: your AI assistant lives where your work lives. That setup is convenient, and this week it became a single point of failure.
The point that matters most to me is number six. Companies that hooked ChatGPT or Claude into their Microsoft 365 data through connectors found out those tools rely on Microsoft’s backend too. Running more than one AI vendor didn’t protect them, because the data layer underneath was the same.
This is the dependency problem enterprise IT has always had, and AI makes it worse. A dead inbox costs you an afternoon. When AI agents have been taking over workflows like triaging email, drafting replies, summarizing meetings and searching files, a backend failure stops all of that work at once.
📉 Context: How We Got Here
Cloud outages aren’t new. Microsoft, Amazon and Google have each had major incidents that took down services for millions of people. What’s changed is the stakes.
Copilot used to be a tool you tried out. Now it’s built into the products companies pay for and the processes they run every day. Vendors sell AI agents as always-on digital workers. This outage shows that those workers are only as reliable as the infrastructure under them.
🛡️ Tactical Recommendations
If your team runs on Microsoft 365, here’s what to do:
- Map your AI dependencies. List every AI tool and agent that touches Microsoft data, whether directly or through a connector.
- Write manual fallbacks. Any workflow an agent handles needs a documented human backup.
- Watch the service health dashboard. Check advisory MO1550821 in the Microsoft 365 admin center for official updates.
- Don’t assume recovery is permanent. Recovery has already started and failed in some regions. Confirm services are stable before you restart automated jobs.
- Ask your vendors hard questions. Find out how your AI providers handle an upstream data outage and what they’ll tell you when one happens.
🔭 What Comes Next
Expect a detailed post-incident review from Microsoft. Expect enterprise customers to push back on service-level agreements too, especially for Copilot add-on licenses that cost real money per seat.
The bigger question is about the agent era. More AI runs on a few cloud platforms every month, and outages like this one will affect more people each time. Resilience planning is now part of an AI strategy, not something to bolt on later.
The Information has more details on Microsoft’s recovery efforts.