The timeline Microsoft published
Status checked through 11:57 p.m. ET on November 12 (04:57 UTC, November 13).
The failure started smaller than it looked. Microsoft's own record puts the first break at 11:17 a.m. Eastern on Thursday, November 12, when users could not connect to Exchange Online mailboxes through any of the usual routes: Outlook on the web, desktop Outlook, mobile sync or the REST interface. Microsoft opened the broader incident MO1550821 at 1:26 p.m. ET, listing impacts across multiple Microsoft 365 services, because so much of the suite leans on Exchange. Teams calendars and invites failed; Microsoft listed core chat, channel messages and joining meetings as not impacted. Copilot features that read mail, calendar and files went with them.
Scale is hard to state cleanly, and the available number needs its label. Downdetector counted 138,058 Outlook problem reports around 1 p.m. Eastern, falling to about 34,490 by the evening. Those are user-submitted problem reports, not a count of unique users, affected customers or tenants. In the records reviewed through that cutoff, Microsoft published no affected-user figure.
An explanation that moved during the day
Read Microsoft's updates in order and the account changes. At 2:06 p.m. it paused all deployments and configuration changes while it reviewed recent changes for a correlation. At 2:57 p.m. it said it had not identified any recent change that correlated with the start of impact. At 8:06 p.m. it said it had begun reverting a recent network configuration change that does correlate with the start of impact, and that a process responsible for establishing secure connections was becoming overwhelmed on the affected servers.
That is an investigation moving, not a contradiction. But a correlation identified at 8:06 p.m. is not a root cause, and the incident record still lists the root cause as N/A. The honest summary at midnight is narrow: a large number of Microsoft's own servers became unavailable after failing automated health checks, capacity fell, load concentrated on what remained, and servers brought back kept failing again once they received traffic.
Why an email outage reached the AI tools
Reuters reported that the disruption also hit the Microsoft connectors used by OpenAI's and Anthropic's AI services. That detail is the part worth keeping. The connectors are how those AI services reach a customer's mailbox, calendar and files. An otherwise functioning AI service can lose access to Microsoft-hosted work data when its connectors are disrupted, however healthy the model itself is. The outage also had at least one dated business consequence: Workhorse Group moved its third-quarter results announcement from November 12 to November 16, citing the disruption.
What to do with this
If your organization runs on Microsoft 365, the useful records are the two incident IDs, EX1550543 and MO1550821, in the admin center's Service health page. That feed, not a status aggregator, is where the recovery declaration and the post-incident review will appear. Microsoft's service description commits it to a preliminary review within 48 hours of an incident being resolved and a final review within five business days, and this incident was not resolved at the last update checked, 11:57 p.m. ET on November 12. Until that review lands, treat any single-cause account of Thursday, including confident ones, as ahead of the evidence.
Sources
- [1] Microsoft Service health incidents EX1550543 and MO1550821, timeline as mirrored by the University of Pennsylvania (status checked to 04:57 UTC, November 13, 2026)Read source
- [2] Microsoft Service health timeline for EX1550543 and MO1550821, reproduced in full by EpiFiveRead source
- [3] Reuters, "Microsoft outage drags on for hours, disrupting Outlook, Teams for thousands," via BusinessWorld, November 13, 2026 (Downdetector peak and evening figures, OpenAI/Anthropic connector impact, Workhorse Group postponement)Read source
- [4] Microsoft Learn, Service health and continuity (post-incident review timing)Read source
- [5] Microsoft 365 public status pageRead source