What happened to Claude?
Anthropic’s incident history provides a more precise timeline than a single “Claude is down” label:
| Time on November 13, UTC | Anthropic’s reported status |
|---|---|
| 09:58 | Investigation began into authentication problems affecting Claude.ai and Claude Code. |
| 11:05 | A fix had been deployed, with recovery being monitored. |
| 11:58 | The earlier incident was marked resolved. |
| 13:09 | A separate investigation began into degraded performance affecting Claude.ai, the Claude API and Claude Code. |
| 15:38 | The second incident was marked resolved. |
These were two incident records, not one uninterrupted outage. The earlier record also lists Claude Console and Claude Cowork among affected components.
Gulf News reported an increase in Downdetector complaints, including reports from India. It also said the extent of disruption in the United Arab Emirates had not been confirmed and that Anthropic had not issued a UAE-specific advisory. Those reports should not be read as a count of affected customers or proof of a country-wide outage.
Recovery came with qualifications
Shortly before closing the second incident, Anthropic said error levels had returned to normal for Claude.ai, its mobile apps, the API and Claude Code. However, its 15:33 UTC recovery update still identified operational limitations:
- Adding a new connector was temporarily unavailable.
- Scheduled tasks and routines could run more than two hours late, with some runs skipped.
- Claude Security scans, reviews and project setup remained delayed.
- Existing Docs and Sheets connectors were working again, while Claude Managed Agents and Claude Meetings were recovering.
The incident was closed five minutes later, without a detailed follow-up addressing each limitation individually. For teams relying on scheduled work, checking whether a particular job completed is therefore more useful than assuming a resolved banner guarantees every missed task has caught up.
Anthropic’s overall dashboard also continued to show degraded performance for Claude.ai and Claude Cowork, while listing Claude Code, the API and Console as operational. That component snapshot sits alongside the resolved incident records; it is not evidence that every Claude feature was continuously unavailable—or fully restored at the same moment.
The Microsoft 365 problem was separate
The Microsoft-linked incident began on November 12, before Friday’s Claude authentication failures. Anthropic initially reported slow responses and errors involving mail, SharePoint, Teams chat and calendar requests through its Microsoft 365 connector. New connections could also fail to complete. It subsequently attributed that incident to a Microsoft 365 service disruption.
Claude for Outlook was affected as well: mail and calendar actions could fail, and some users could not open the add-in. In its latest posted update, at 15:40 UTC on November 13, Anthropic reported recovery for both the connector and Claude for Outlook, but said some SharePoint searches still returned errors. Microsoft’s incident remained open, and Anthropic warned that errors could return.
There is no basis in these incident notices to blame Microsoft for Claude’s separate sign-in and performance failures. Anthropic tracked the connector problem independently and attributed that specific problem—not the other two incidents—to Microsoft 365. Coincidence on the clock is not a root-cause analysis.
What users and IT teams should check
The practical takeaway is to identify which operation is failing, rather than treating every error as the same outage:
- Cannot sign in to Claude: Compare the failure with Anthropic’s authentication incident history.
- Claude opens, but cannot retrieve Microsoft 365 information: Check the separate Microsoft 365 connector incident.
- Claude for Outlook will not open or perform an action: Include the add-in’s recovery status in the investigation.
- A scheduled workflow appears unfinished: Verify its actual completion before deciding whether to run it again.
These are triage recommendations based on the different failure paths Anthropic documented, not a claim that every affected account experienced all of them.
The useful distinction is straightforward: core Claude availability, connected Microsoft 365 access and background-task completion are separate things to verify. Friday’s incident records show why a working chat window alone is not a complete health check for an AI-assisted business workflow.