AI service incidents · Daily incidents

AI incidents: Microsoft 365 disrupts Copilot, Claude and ChatGPT connectors — 13 November 2026

Microsoft 365 Copilot and connectors in Claude and ChatGPT had disruptions. OpenAI reported recovery for Sites publishing and its image API.

Publisher disclosure: MUVO offers competing music, remix, and video tools. This report summarizes public sources and provider documentation; it is not an independent product test or a claim that MUVO is free of similar issues. Read our reporting method.

Recovery update, 14 November

Microsoft marked its Microsoft 365 incident restored in the tenant update at 07:17 Istanbul on 14 November, while saying it was still restoring service for remaining customers and regions. Microsoft attributed the incident to an issue introduced during an earlier network capacity upgrade that surfaced after a later network change. The team said it would publish a full post-incident report. The tenant record lists a formal impact window from 13:05 on 12 November to 19:10 on 13 November. Those fields describe Microsoft’s incident record. The connector operators reported their own recovery timelines. Microsoft tenant notice reproduced by Penn IT.

Anthropic reported errors for many connector and Claude for Outlook requests from 19:03 on 12 November to 19:40 on 13 November. It began monitoring at 18:40 on 13 November and posted resolution at 00:49 on 14 November, with occasional delays still possible. Anthropic incident notice.

OpenAI reported signs of recovery at 19:20 on 13 November and marked its Microsoft 365 connector incident resolved at 03:18 on 14 November. OpenAI connector notice.

Incident summary at the original cutoff

Microsoft 365 work features remained degraded at the research cutoff of 12:10 Istanbul on 13 November. Microsoft reported trouble loading or getting responses from work Copilot Chat and accessing work data. Anthropic and OpenAI separately reported failures in their Microsoft 365 connections. These notices describe specific affected features, not a shutdown of either assistant.

Microsoft 365: Exchange Online disruption reaches work Copilot

Microsoft says the disruption affects features that rely on Exchange Online. Work Copilot Chat may fail to load or respond, and Copilot may have trouble reaching work mail, calendar items and files. Microsoft says most core Teams calls and chats remained functional, while Exchange-dependent calendar and some Copilot actions could fail. This notice covers Microsoft 365 for business and enterprise. Microsoft service health.

The Microsoft tenant notice MO1550821 lists its start as 12 November, 21:26 Istanbul. At 10:29 on 13 November, Microsoft’s public notice reported improved server stability on part of its infrastructure after a patch to an authentication and routing component. The 11:09 tenant update said Microsoft was expanding the patch to infrastructure serving regions in or nearing peak business hours. These are mitigation updates; a confirmed end to impact and a final root cause were unavailable at cutoff. Microsoft tenant notice reproduced by Penn IT.

Claude: mail, calendar and search tools fail

Anthropic acknowledged connector errors at 20:52 on 12 November and identified an upstream issue at 21:34. At 01:23 on 13 November, it attributed the disruption to Microsoft 365 and said most requests to mail, calendar and search through the connector were still failing. Claude for Outlook also remained affected. Anthropic incident notice.

ChatGPT: new Microsoft connections and tool calls may fail

OpenAI acknowledged the connector incident at 21:29 on 12 November. Its notice says new connections may fail and existing tools may not work as expected. The affected ChatGPT component is Connectors/Apps. The notice remained under investigation at cutoff. OpenAI connector notice.

OpenAI: Sites publishing and image API recover

Sites: OpenAI recorded degradation for ChatGPT Sites publishing beginning at 04:17 on 13 November, acknowledged elevated errors at 06:59, and posted resolution at 08:55. The scope is publishing; the notice does not establish that already published sites were inaccessible. Sites incident.

Image API: OpenAI marked the GPT-image-2.5 and 2.6 Sunburst and Flare incident resolved at 12:25 on 12 November, after reporting mitigation and monitoring at 12:18. We added the recovery to the 12 November report. Image API notice.

xAI, Fireworks and Bolt: separate resolved incidents

The xAI status feed records request errors including purchases and billing actions in incident ps_94275750e294382d. The first update named cursor.com; later updates included Grok Bot. The first notice was at 23:08 on 12 November, monitoring began at 23:43, and resolution was posted at 00:39 on 13 November. These notice times do not establish the duration of an individual user’s impact.

Fireworks posted these component intervals for models served by Fireworks. The DeepSeek intervals overlap and should not be added together. They do not establish an outage on DeepSeek’s own API.

Fireworks model Start (Istanbul) End (Istanbul)
Service Degradation for GLM 5.3 Fast 12 Nov 18:05:27 12 Nov 18:19:48
Service Degradation for DeepSeek V4.1 Flash 1017 12 Nov 23:33:24 13 Nov 00:26:40
Service Degradation for DeepSeek V4.1 Flash 1017 12 Nov 23:39:27 12 Nov 23:51:35
Service Degradation for DeepSeek V4.1 Flash 1017 13 Nov 02:06:27 13 Nov 02:20:41
Service Degradation for GLM 5.3 13 Nov 03:18:27 13 Nov 03:24:49
Service Degradation for GLM 5.3 13 Nov 03:57:26 13 Nov 04:32:39
Service Degradation for GLM 5.3 Fast 13 Nov 04:14:27 13 Nov 04:24:35
Service Degradation for GLM 5.3 13 Nov 05:11:34 13 Nov 05:13:21
Service Degradation for GLM 5.3 Fast 13 Nov 05:38:27 13 Nov 05:44:51

Bolt recorded three distinct resolved degradations for its Prompting feature. Its notices did not name a cause.

Bolt component Start (Istanbul) End (Istanbul)
Prompting 12 Nov 17:27:02 12 Nov 17:28:47
Prompting 13 Nov 01:19:30 13 Nov 01:31:34
Prompting 13 Nov 11:39:26 13 Nov 11:40:11

What affected users can check

If a connector cannot retrieve mail, calendar items or work files, verify any resulting answer against the original record once access returns. Before retrying an action that may write to another service, check whether the original action already went through. For workplace impact and updates, use your tenant’s service health view or Microsoft’s public notice.

Sources

The report uses Microsoft service health, Anthropic’s incident history, OpenAI’s incident history, and the individual notices linked above. All displayed times use Istanbul time (UTC+3). Where impact boundaries are unknown, notice times are not estimates of downtime. The public review scan did not supply corroborated customer allegations for publication; several review sites restricted access. See reporting methodology.

Update timeline