News · 2026-11-13

Microsoft 365's 18-hour outage knocked out Claude's and ChatGPT's connectors while the models themselves stayed up

Microsoft's cloud outage that began around 1:26 p.m. Eastern on November 12, 2026 disrupted Exchange Online, Teams, SharePoint Online, Microsoft 365 Copilot and Defender XDR for at least 18 hours, according to the BBC. The AI angle is the integration layer: Claude's Microsoft 365 connector failed for about five hours in Anthropic's status feed, and OpenAI said ChatGPT's Microsoft connectors were not working as expected. The models stayed up; the pipes to work data did not.

Key facts

• Start and length: Microsoft dates the incident from Thursday 1:26 p.m. ET; the BBC's live page says it ran "at least 18 hours," and BleepingComputer's last update at 8:05 a.m. ET on November 13 reported "encouraging recovery progress" with intermittent issues possible.
• Scale: CNN says Downdetector reports "reached 74 countries and territories in about an hour"; Microsoft says more than 430 million people use its 365 apps.
• Cause so far: a number of servers unexpectedly unavailable, straining the rest. No final root cause was found.
• Claude: Anthropic's status feed lists the connector problem from 17:52 UTC to at least 22:23 UTC with no resolved entry.

What Microsoft says happened

Microsoft first blamed elevated infrastructure load across multiple regions, then posted on X: "We determined the elevated load isn't driven by an increase in traffic, but by a number of servers being unexpectedly unavailable; straining the remaining infrastructure." It paused all deployments, redistributed traffic and restarted servers. A later update said recovery had not given the expected relief because "a core process on affected servers was becoming overwhelmed, resulting in a loss of capacity," and that it was weighing a revert of "a small number of recent configuration updates." Charles Lamanna, executive vice president for Copilot, Agents and Platform, said there was no indication of "malicious or cyber activity." Microsoft extended its email timeout window to 48 hours so delayed messages would not bounce. An earlier Teams incident that morning, blamed on "resource contention," has not been linked to this one.

Real-world impact

Several UK police forces said online forms were disrupted and told people to phone. NHS email was disrupted and some appointments were cancelled, the BBC reports. The US House of Representatives and city governments reported email problems, and a Colorado sheriff's office said bail payments were delayed. Workhorse Group moved an earnings call to November 16 and Richemont delayed results.

The AI-specific damage

Microsoft's own list of affected services includes Microsoft 365 Copilot. That Copilot has nothing to read when Exchange, SharePoint and Teams fail is an inference, not a published statement. The clearer evidence is on the connector side. Anthropic's status feed titled the first Claude incident "Elevated errors affecting the Microsoft 365 connector." At 17:52 UTC requests to mail, SharePoint, Teams chat and calendar through the connector could fail; at 18:34 UTC it was identified as an upstream provider issue; at 20:28 UTC Claude for Outlook was also affected; at 22:23 UTC the feed said the problem was "caused by a Microsoft 365 service incident," with the connector still failing for most requests. CNN separately reports OpenAI saying some tools, including ChatGPT, were not working "as expected" because of Microsoft connectors; OpenAI's own status page was not checked. See our lesson on tool use and function calling for why connectors are a dependency chain.

A separate Claude incident on November 13 involved users authenticating to claude.ai and Claude Code; a fix deployed at 11:05 UTC and it was resolved at 11:58 UTC. Anthropic gave no cause and no link to Microsoft is established.

Reception and caveats

The Hacker News status-page thread reached 1,007 points and 850 comments; other threads reached 402 and 386. A BleepingComputer commenter blamed layoffs and AI-generated code, but that is speculation, and nothing Microsoft published points to AI tooling. The final resolution time, root cause and a post-incident review are not yet published, so check Microsoft's status page before relying on any duration.

Why it matters

An AI agent that reads your mail is only as available as the mail service. Multiple assistants inherited one vendor's failure without any model downtime, which is a concrete argument for graceful fallbacks when a connector's upstream disappears.

Primary source, verified: read the paper →

Ground Truth. (2026, November 13). Microsoft 365's 18-hour outage knocked out Claude's and ChatGPT's connectors while the models themselves stayed up. Ground Truth. https://groundtruth.day/news/microsoft-365-outage-took-ai-connectors-down-while-the-models-stayed-up.html

@misc{groundtruth:microsoft-365-outage-took-ai-connectors-down-while-the-models-stayed-up,
  title  = {Microsoft 365's 18-hour outage knocked out Claude's and ChatGPT's connectors while the models themselves stayed up},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {nov},
  url    = {https://groundtruth.day/news/microsoft-365-outage-took-ai-connectors-down-while-the-models-stayed-up.html}
}

Topics: microsoft · outage · copilot · claude · chatgpt · infrastructure · connectors