A worldwide Microsoft 365 outage that began on Thursday, November 12, was still disrupting Outlook, Teams, SharePoint and Microsoft 365 Copilot on Friday morning, more than 15 hours after the first reports. Microsoft has not said when service will be fully restored. The failure has also broken the Microsoft 365 connectors that Anthropic’s Claude and OpenAI’s ChatGPT use to read mail, calendars and files, which shows how much workplace AI depends on one company’s servers.
This story describes the situation as of 07:45 UTC on November 13. The incident is still open, and details may change.
What is down
User reports started at about 11:15 a.m. Eastern time on Thursday, and Microsoft’s status account acknowledged problems with Exchange Online mailboxes “in the Americas” shortly before 1 p.m. The company later said users in any region could be affected. Administrators can follow the incident under EX1550543 and MO1550821 in the Microsoft 365 admin center, according to BleepingComputer and ITdaily.
Exchange Online, the mail service behind Outlook for businesses, is at the centre of the failure. Microsoft’s advisory also names Teams, SharePoint Online, Microsoft 365 Copilot and Defender XDR, and the list published by Fox Business adds Intune, Purview, Planner, To Do, Power BI and Windows 365. Downdetector logged more than 130,000 Outlook problem reports early on Thursday afternoon, and ITdaily reports that they came from 74 countries and territories.
What Microsoft says went wrong
Microsoft’s explanation has changed several times. It first blamed “elevated load across multiple regions”. By Thursday evening it said the load was a symptom: a large number of servers had failed automated health checks and dropped out of service, which pushed their traffic onto the remaining machines.
Late on Thursday, Microsoft said that “a process on the affected servers responsible for establishing secure connections becomes overwhelmed, causing the servers to stop processing requests”. It also began reverting a recent network configuration change that, in its words, “correlates with the start of impact”. Microsoft has not confirmed that this change caused the outage.
Recovery has been uneven. Microsoft paused all deployments, moved traffic away from the affected infrastructure and restarted servers, then warned that restarted servers “can become unresponsive again once they receive traffic”. Around 2:30 a.m. Eastern on Friday, it reported “positive results for server stability” after patching what it called an upstream authentication and routing configuration component on part of its infrastructure. It is also adding capacity with extra virtual machines.
The company says it is “not aware of any malicious activity” behind the incident and has promised a post-incident report with the root cause once service is restored.
How the outage reaches AI assistants
Microsoft 365 Copilot is the most direct casualty. Microsoft’s advisory says Copilot Chat may fail to load or respond in the web and desktop apps, and ITdaily reports that the assistant cannot reach the email, calendar and files it normally works from. Microsoft 365 Copilot answers questions by retrieving a user’s own work data, so when Exchange Online stops answering, most of its usefulness goes with it.
Rival assistants that plug into Microsoft 365 are failing for the same reason. Anthropic opened an incident at 17:52 UTC on Thursday for the Microsoft 365 connector in Claude. Its status page says mail, calendar and search tools are “failing for most requests”, that the Claude for Outlook add-in is affected, and that the cause is the Microsoft 365 service incident. OpenAI opened its own incident at 18:29 UTC, saying new connections to the Microsoft connectors in ChatGPT “may be broken” and tools are not working as expected. In the latest updates we saw, neither company had marked its incident as resolved.
The AI models themselves are not affected. Claude and ChatGPT still answer ordinary prompts; what has failed is their link to data held in Microsoft’s cloud.
Who is affected
Governments were among the first to report problems. The US House of Representatives said the outage is affecting its users, Japan’s deputy chief cabinet secretary said government information systems were hit, and New Zealand’s foreign ministry said email to its embassies and consulates worldwide was disrupted, according to ABC News in Australia. New York City’s emergency management department warned residents of delays in reaching city agencies by email.
Companies have felt it too. CNBC reports that Richemont slightly delayed its earnings release, Workhorse Group moved its third-quarter results to November 16 and Air India temporarily rerouted emails without any effect on flights. Canberra Health Services said email, voicemail messages and some faxes were disrupted, with no known impact on clinical care.
Amazon said its own services were “operating normally” after some users reported problems with Amazon Web Services on Downdetector.
What it means for companies building on Microsoft 365
For organisations that have deployed AI assistants, the outage is a practical test of where those tools get their data. An assistant that depends on live access to a mailbox or a SharePoint library stops being useful when that service is down, whichever company makes the model. Buying AI from a second vendor does not help if both read from the same Microsoft 365 tenant.
The episode also comes two weeks after Microsoft put a dollar figure on Azure for the first time, a reminder of how much business now runs through its cloud. Teams that are deciding how much to keep on their own hardware can find the trade-offs in our comparison of cloud APIs and local models.
It is not Microsoft’s first long outage this year. ITdaily notes that Microsoft 365 was down in North America for almost ten hours in January.
What to watch next
- Microsoft’s confirmation that service is fully restored, and how long the backlog of delayed email takes to clear.
- The post-incident report, in particular whether the reverted network configuration change was the trigger.
- When Anthropic and OpenAI close their connector incidents.
- Whether large customers or regulators ask Microsoft for changes after a failure that reached several governments.
Key takeaways
- A Microsoft 365 outage that started on November 12 was still unresolved on the morning of November 13.
- Exchange Online is the core of the failure; Teams, SharePoint, Copilot and Defender are affected through it.
- Microsoft points to servers that fail health checks and a secure-connection process that becomes overwhelmed, but has not confirmed a root cause.
- Microsoft 365 Copilot and the Microsoft connectors for Claude and ChatGPT are degraded because they depend on the same services.
- Microsoft says it sees no sign of malicious activity.
Frequently asked questions
Is the Microsoft 365 outage fixed?
Not as of 07:45 UTC on November 13. Microsoft reported positive results from a patch on part of its infrastructure but gave no time for full recovery. Its service health page at status.cloud.microsoft has the current state.
What caused the Microsoft outage?
Microsoft says servers failed automated health checks and that a process responsible for secure connections becomes overwhelmed, so servers stop processing requests. It is reverting a recent network configuration change but has not confirmed a root cause, and it says it has seen no malicious activity.
Does the outage affect Copilot, Claude and ChatGPT?
Microsoft 365 Copilot may fail to load or to reach email, calendar and files. Claude and ChatGPT still work for normal prompts, but their Microsoft 365 connectors, which read mail, calendars and files, are failing during the incident.
