Microsoft 365’s outage hits Copilot and the Claude and ChatGPT connectors. What to do

What’s down

On November 12, 2026, Microsoft 365 started failing for people around the world. Microsoft first reported trouble reaching Exchange Online mailboxes in the Americas, under incident EX1550543, then opened a wider one, MO1550821, which it dates from 18:26 UTC (1:26 p.m. Eastern): “Users may be unable to connect to multiple Microsoft 365 services.” Al Jazeera reports that Downdetector logged about 130,000 Outlook problem reports shortly after 1 p.m. Eastern.

What breaks, by Microsoft’s status page, is the features “that interact with Exchange Online and email services”: mail, calendars and contacts, and everything built on them in Outlook, Teams, SharePoint, Copilot and a long list of admin and security tools. Its consumer listing names Outlook.com, OneDrive, OneNote, Microsoft To Do, Teams and Microsoft 365 for the web.

The AI assistants that read your mail fail with it. Here is each, as its maker described it on the morning of November 13 (UTC):

Assistant	What fails	Status
Microsoft 365 Copilot	Copilot Chat may not load or answer, and Copilot “may be unable to access work data such as email, calendar, and files”; in Teams, Copilot fails on Outlook, calendar and files	Degraded, part of MO1550821
Claude’s Microsoft 365 connector	Mail, calendar and search tools “failing for most requests”; new connections “may not complete”	Open on Anthropic’s status page since 17:52 UTC, November 12
Claude for Outlook	“Mail and calendar actions may fail, and some users may not be able to open the add-in”	Open, the same incident
ChatGPT’s Microsoft apps	“New connections may be broken and tools are not working as expected”	Under investigation on OpenAI’s status page since 18:29 UTC, November 12
Agents and scripts on Microsoft Graph	Requests for “mail, calendar, and contacts, may time out”	Degraded, part of MO1550821
Microsoft Copilot, the consumer app	Nothing reported	Operational
GitHub Copilot	Nothing reported	Operational

Anthropic’s incident lists claude.ai and Claude Cowork, and its last update, at 22:23 UTC on November 12, says: “This issue is caused by a Microsoft 365 service incident.” It doesn’t say whether personal Microsoft accounts, which Claude’s Free, Pro, Max and Ultra plans can also connect for Outlook and OneDrive, are affected. OpenAI’s incident names “Microsoft connectors” without listing them, and has had no update since its first; ChatGPT’s Microsoft apps are Outlook Email, Outlook Calendar, Microsoft Teams and SharePoint, by its help center. Neither status page shows another open incident, so nothing else at either company is reported down.

GitHub’s status page shows Copilot operational. GitHub did report delayed email “related to an upstream provider degradation” on November 12, without naming the provider, and resolved it later that day.

What Microsoft says caused it

Microsoft hasn’t published a root cause, and its explanation has moved as it worked. From its advisory, in UTC:

Time	What Microsoft said
Nov 12, 18:27	The infrastructure is “experiencing elevated load across multiple regions”
Nov 12, 19:30	The load “isn’t driven by an overall increase in traffic, but by a large number of servers being unexpectedly unavailable”; all deployments and configuration changes paused
Nov 12, 22:03	“Our actions taken thus far have not provided the anticipated relief”; “a core process on the affected servers is becoming overwhelmed”
Nov 13, 00:00	Servers returned to service “can become unresponsive again once they begin receiving traffic”
Nov 13, 01:12	“A process on the affected servers responsible for establishing secure connections becomes overwhelmed”; reverting “a recent network configuration change that correlates with the start of impact”
Nov 13, 04:12	Partial recovery in some regions, but “these actions aren’t resulting in sustained recovery”
Nov 13, 07:24	“Positive results for server stability” from “an upstream authentication/routing configuration component patch” on part of the infrastructure
Nov 13, 08:09	Widening the patch, starting with “regions currently within or entering peak business hours”
Nov 13, 09:03	“Early signs of stabilization, but service capacity has not yet returned to normal levels”; the email timeout window extended “to 48 hours to help protect delayed messages”

Microsoft says “we are not aware of any malicious activity that is causing this incident” and that, “once mitigated,” it will give the root cause in a Post-Incident Report. The full advisory sits in the Microsoft 365 admin center, which Microsoft says may not load for some admins; Cloud Admin Hub keeps a public copy. Earlier on November 12, from 9:45 to 10:53 a.m. Eastern, a separate Teams incident, TM1550488, hit North and South America, which Microsoft put down to resource contention, BleepingComputer reports. Microsoft hasn’t said whether the two are related.

What still works

Microsoft says “elements of those services that do not depend on interaction with Exchange Online remain functional.” In Teams, “most of the core functionality including joining calls, meetings, group chats, and channel messages are not impacted.” In SharePoint, “accessing files stored directly in SharePoint Online remains functional.”

Recovery is uneven. Computerworld reports that Microsoft is restricting some traffic and restoring connections gradually, and Al Jazeera that problems were reported in Asia and Australia on Friday morning. Whether your assistant works may depend on where your organization’s mail is served from.

What to do until it’s fixed

1. Leave your connections alone. Anthropic and OpenAI both say new connections may fail, so disconnecting Microsoft 365 to fix it can leave you with no connection until the outage ends. Microsoft’s advice for its own Power Apps mobile apps runs the same way: people already signed in “can potentially avoid impact by not signing out”.
2. Don’t trust an empty answer. While the mail and search tools fail for most requests, an assistant that says it found no matching email may not have reached your mailbox at all. Ask again after recovery before acting on “nothing found.”
3. Bring the file to the assistant. Files in SharePoint still open, so download the document and attach it to Claude or ChatGPT, where your organization’s policies allow it, instead of asking a connector to find it.
4. Check what your agents did. With write actions on, Claude’s connector can send mail and change calendar events, and Microsoft says “email delivery may be delayed”; on November 13 it said it was “extending the email timeout window to 48 hours to help protect delayed messages.” After recovery, check Sent Items and your calendar for anything an assistant did during the outage, and rerun scheduled tasks that read your mail.
5. Watch the right status page. Microsoft’s is status.cloud.microsoft, which Anthropic also points to, and its Microsoft 365 Status account on X; admins can follow MO1550821 and EX1550543 in the admin center. For the assistants, see Anthropic’s and OpenAI’s incident pages above.

What Microsoft hasn’t said

• When service will be fully restored.
• The root cause, beyond the overwhelmed process and the network change that “correlates” with the start.
• Whether any email was lost. It says delivery “may be delayed” and that it is working “to preserve delayed mail during recovery.”
• Whether the morning’s Teams incident was related.
• Anything about service credits: its updates don’t mention them.

We’ll update this memo when Microsoft marks MO1550821 resolved or publishes its Post-Incident Report.