Microsoft 365 outage takes Exchange Online, Copilot, and rival AI mail connectors offline for more than 18 hours

Documented: A Microsoft 365 failure that began at 16:17 UTC on 12 November 2026 left Exchange Online mailboxes unreachable through Outlook on the web, desktop, mobile, and REST connections, with delayed email delivery. Microsoft’s own service notices widened the incident beyond the Americas to any region and listed knock-on effects wherever a product had to touch mail, calendar, or contacts. Teams calls and group chats largely continued. Calendar edits, meeting invites, and anything that queried a mailbox did not.

Microsoft revised its account of the cause in public status updates. It first pointed to elevated load, then said the load was not driven by a surge in customer traffic. Servers were failing automated health checks and dropping out of rotation, dumping work onto the machines still running. Restored servers became unresponsive again once traffic returned. Engineers paused deployments, patched an authentication and routing component, and extended the email timeout window to 48 hours so delayed messages would not be discarded. By Friday morning the company said queues were significantly reduced and approaching normal, with intermittent access problems still possible. It said it had seen no evidence of malicious activity and promised a post-incident report with root cause and defense-in-depth actions once mitigation was complete. Charles Lamanna, executive vice president for Copilot, agents, and platform, posted that there was “no indication this is the result of malicious or cyber activity.”

The AI layer failed with the inbox. Microsoft’s status notice said Copilot Chat might not load and that Copilot experiences might be unable to access work data such as email, calendar, and files. Anthropic and OpenAI separately reported that their Microsoft 365 connectors—the path Claude and ChatGPT use to reach a customer’s mail, calendar, and related search—were failing for most requests. ChatGPT itself was not down as a product; the connector was. That is the operational fact under the “agentic economy” marketing: an assistant that reads the office still depends on Exchange Online and Microsoft Graph staying up.

Real-world traction is not speculative. Downdetector logged on the order of 130,000 U.S. Outlook problem reports at the peak—user reports, not a census of affected tenants. Microsoft has said paid Microsoft 365 commercial subscriptions exceeded 450 million. In Australia, Home Affairs Minister Tony Burke said a formal agreement with Microsoft was activated so Australian interests would be prioritized in restoration; ASIO and the Australian Federal Police confirmed they were affected at least in part. Queensland Health stood up an emergency command center while saying patient care and Triple Zero were not disrupted. The Australian Medical Association reported some clinics falling back to paper prescriptions because multifactor login depended on Microsoft-backed email. In Japan, Deputy Chief Cabinet Secretary Masanao Ozaki said the government’s shared Government Solution Service was hit and staff were producing paper copies; the Asahi Shimbun reported the Cabinet Office posting notices on a press-room board and a Defense Ministry official saying an emergency press release might have to be posted on paper inside the ministry. Nikkei reported market participants turning to fax because confirmation email had stopped, and delays in publishing some investment-trust reference prices. Prime Minister Sanae Takaichi instructed ministries to keep non-email contact paths open; the chief cabinet secretary said medical records and financial services were not, at that briefing, showing disruption.

Claimed, not independently verified here: ZeroHedge’s framing that traders “reached for fax machines” and that Japan’s finance ministry “pinned paper notices to the wall” compresses several Japanese reports. What is documented is narrower—fax and telephone as market workarounds, paper copies and bulletin-board notices inside parts of the Japanese government, not a single finance-ministry wall notice confirmed in English-language wire copy.

Speculated, and not supported by Microsoft’s notices: that the outage was a cyberattack, or that a “rogue” internal AI caused it. Both circulated on social feeds. Lamanna and Microsoft’s status page contradicted the attack theory. Root cause, as of the Friday updates cited above, had not been published.

The dependency is the story the capability coverage usually skips. One mail-and-calendar backend took Microsoft’s own agent and two competitors’ office connectors down together. Security tooling that reads the same mail path—Defender for Office 365 Safe Links, parts of Purview and Defender XDR—was listed as delayed or failing even while Azure itself was reported clear. An agent that cannot see the mailbox cannot do the job the product page describes, regardless of model quality. Concentration risk here is not abstract: governments invoked priority-restoration agreements, clinics reverted to paper for authentication-gated prescribing, and a market that still keeps fax as a fallback used it. Whether this becomes Microsoft’s longest 365 outage is a comparison some outlets drew against a 2022 incident of roughly 16 hours; that ranking is a claim about duration, not a finding in a post-incident report that had not yet been issued.