Did the Claude outage affect Microsoft 365 Copilot?

We have no evidence that it did. Microsoft's subprocessor documentation does not say how an Anthropic incident would show up in Copilot, and Anthropic's incidents listed claude.ai, the Claude API, Claude Code and Claude Cowork as the affected components. In ProIT NW's own Microsoft 365 tenant, the Service health issue history showed no Copilot incident that started on October 23 or 24, 2026. That is one tenant's view of resolved items, not a statement about every tenant — check your own Service health.

How long was Claude down?

Anthropic's status page gives three impact windows on October 23, 2026 (Pacific): 4:28–4:59 PM for Claude Fable 5.1, Sonnet 5.5 and Haiku 5.5 plus fast-mode requests to Opus 5.5, rated critical; 5:23–5:34 PM for Claude Opus 5.5; and 10:00–10:35 PM for other Claude models. That is about 77 minutes of impact across the evening, in three separate incidents.

Did the problems continue after October 23?

Yes, briefly. Anthropic's status page recorded two more short, resolved incidents over the next two days: elevated errors for Claude Sonnet 5.5 and other models on October 24 (11:05–11:35 PM PT, about 30 minutes) and for Claude Fable 5.1 on October 25 (1:08–1:29 AM PT, about 21 minutes). Both hit the same components as the 23rd. We have not seen a root cause, and we found no sign either one reached Microsoft 365 Copilot in ProIT NW's own tenant.

Should we drop the AI vendor that had the outage?

One evening of incidents is not, on its own, a reason to switch. Every cloud service has outages — Microsoft 365 and Azure included. The useful response is to know where AI sits in your critical processes, have a fallback for those steps, and make sure any alternate model you fall back to is one your data-governance rules already allow.

Blog · 7 min read · October 24, 2026

Claude Had Three Incidents in One Evening: Plan for It

Anthropic's status page recorded three Claude incidents on October 23, 2026 (Pacific): 4:28–4:59 PM (critical), 5:23–5:34 PM and 10:00–10:35 PM — about 77 minutes of impact in total.

ProIT NW, a senior-led Microsoft consultancy in Seattle serving the regulated mid-market, is writing about this one because AI is no longer a side tool in most offices. On the evening of October 23, 2026, Anthropic's Claude — the model family behind claude.ai and Claude Code, and one of the model options inside Microsoft 365 Copilot — had three separate incidents in about six hours. None was long. Together they are a good prompt to ask a question most firms have not: if the AI stops, which of our processes stop with it?

What we know, and where it comes from. The times and wording below are from Anthropic's public status page. Microsoft's statements are from its Learn documentation. Where we describe our own Microsoft 365 tenant, we say so. We have not seen a root cause from Anthropic.

What happened on October 23

Anthropic's status page recorded three incidents, each affecting claude.ai, the Claude API, Claude Code and Claude Cowork:

4:28–4:59 PM PT — rated critical. "Elevated errors for Claude Fable 5.1, Sonnet 5.5, and Haiku 5.5." Anthropic's second update widened it: "as well as fast mode requests to Opus 5.5." Anthropic later wrote that "Impact occurred from 16:28 PT / 23:28 UTC on October 23 to 16:59 PT / 23:59 UTC on October 23."
5:23–5:34 PM PT — rated major. "Elevated errors for Claude Opus 5.5." Impact lasted about eleven minutes.
10:00–10:35 PM PT — rated major. Elevated errors "on requests to other Claude models," later titled "Elevated errors for Claude Fable 5 and Claude Fable 5.1." Anthropic: "A fix has been deployed for the issue affecting Claude services and we are monitoring for recovery."

That is about 77 minutes of impact, spread across the end of the business day on the West Coast and late evening. For a team that uses Claude for drafting or research, it was an annoyance. For a process that calls the Claude API automatically — a ticket classifier, an intake summarizer, a document pipeline — it could have been a stop, depending on the model it calls.

Then it happened again, on October 24 and 25

The evening of the 23rd was not a one-off. Over the next two days Anthropic's status page recorded two more short incidents, both resolved after a fix and both affecting the same components — claude.ai, the Claude API, Claude Code and Claude Cowork:

October 24, 11:05–11:35 PM PT (06:05–06:35 UTC on the 25th) — elevated errors for Claude Sonnet 5.5 and other models, about 30 minutes.
October 25, 1:08–1:29 AM PT (08:08–08:29 UTC) — elevated errors for Claude Fable 5.1, about 21 minutes.

Short, model-scoped bursts — minutes, not hours — and we still have not seen a root cause from Anthropic, nor any sign the errors reached Microsoft 365 Copilot in our own tenant. The pattern is the point: elevated-error incidents recurred three days running. That does not change the advice below; it is the reason for it. A step that cannot pause for twenty minutes needs a fallback written before the next burst — not a second AI vendor chosen during one.

Update — November 13, 2026: a week of AI-tool outages, and one Anthropic attributes to Microsoft

Five more incidents landed this week. Anthropic prints UTC on its status page; OpenAI prints no timezone, and its times are Pacific (UTC−8 in November).

Microsoft 365 connectors, November 12. Anthropic opened an incident on its Microsoft 365 connector at 17:52 UTC (9:52 AM PT); by 18:34 UTC (10:34 AM PT) mail, calendar and search tools were failing for most requests. Anthropic later wrote: "This issue is caused by a Microsoft 365 service incident." Claude for Outlook was affected too, and as of 15:40 UTC (7:40 AM PT) on November 13 Anthropic was still monitoring, noting that Microsoft's incident remained open. OpenAI's Microsoft 365 connector had errors from 10:29 AM PT the same day ("new connections may be broken"); OpenAI stated no cause. Our post on the Microsoft 365 outage covers the Microsoft side. We are not saying any of this reached Microsoft 365 Copilot.
Claude, November 13. Sign-in failures for claude.ai and Claude Code ran from 09:58 to 11:58 UTC (1:58–3:58 AM PT). A broader incident ran 13:09–15:38 UTC (5:09–7:38 AM PT) across claude.ai, the API, Claude Code and connectors. During it Anthropic said: "Scheduled tasks and routines can run more than two hours late and some runs have been skipped…"
OpenAI, November 11–12. "Unable to start new Workspace Agent runs" was labeled "Full outage" from 2:50 PM PT on November 11 to 12:32 AM PT on November 12, about 9 hours 42 minutes.

The lesson is narrower than "AI goes down." Scheduled agent work can be skipped, not merely delayed, and a skipped run raises no error in the person who was waiting on it. So the continuity question for every scheduled AI job is: what re-runs it, and who notices? Give each one a named owner and a check, such as a confirmation that the expected output exists by a set time. An empty report should count as a failure.

Second, an AI assistant connected to your mailbox fails when the mailbox does. Add every connector to the inventory above, and list the Microsoft 365 service it depends on beside it.

Why a Microsoft shop should care

Many of our clients don't buy from Anthropic directly. That doesn't mean they don't use Claude. Microsoft's documentation on Anthropic as a subprocessor says: "Microsoft enables Anthropic models on by default for most customers in commercial cloud (excluding EU/EFTA and UK)." It lists the places users can reach them: "Microsoft Copilot, Researcher, Copilot Studio, Power Platform, and Copilot in Microsoft 365 apps." Anthropic "has onboarded as a Microsoft subprocessor," and "In Microsoft Copilot (web, desktop, and mobile), UI indicators show when Claude models are in use."

So Claude can be in your Microsoft estate without anyone having signed an Anthropic contract. What Microsoft's subprocessor documentation does not say is how an Anthropic incident would surface inside Copilot. We checked our own tenant: in ProIT NW's Microsoft 365 Service health issue history, no Copilot incident started on October 23 or 24. The only Copilot item in our issue history that was active that day was an unrelated advisory about the account switcher in Copilot Chat. That is one tenant and resolved items only — but it is the honest answer: we found no sign this outage reached Copilot users, and we have not seen Microsoft say either way.

The point is not this particular evening. It is that AI now sits in two places in a typical mid-market firm — the tools staff use directly, and the models embedded in Microsoft 365 — and most continuity plans cover neither.

Where AI is quietly in the critical path

Automations that call a model API. Intake triage, email classification, summarizing scanned documents. If the call fails and there is no retry or fallback, work stops or, worse, passes through unprocessed.
Copilot Studio agents and Power Platform flows that a business team built and IT never reviewed.
Deadline work — a proposal, a regulatory filing, month-end commentary — where the plan quietly assumed the AI drafting step would be there at 4:30 PM.
Developer tooling. Engineering teams on Claude Code or similar lose a working afternoon, not data — but it is still capacity.

What to do this month

Inventory AI in the critical path. List every process where a model call sits between input and output, direct or through Copilot, Copilot Studio or Power Platform. Note who owns it.
Check your Copilot model setting. Find out whether Anthropic models are on in your tenant (Microsoft says on by default for most commercial customers outside the EU/EFTA and UK) and whether that matches your data-governance decision. See our note on the GCC setting that processes data outside FedRAMP if you are in a government cloud.
Subscribe to the status pages that matter. Microsoft 365 Service health for Copilot, and status.claude.com (and any other model vendor) for anything you call directly. An outage you learn about from users costs more than one you learn about from an alert.
Write the fallback before you need it. For each critical step: what happens if the model is unavailable for an hour? Usually the answer is a manual path or a queue, not a second AI vendor.
If you do fall back to another model, pre-approve it. Switching models mid-incident is a data decision. The alternate should already be allowed by your policy — not chosen in a hurry by whoever is on shift.
Log it in your vendor-risk register. Treat model vendors like any other critical SaaS: who they are, what depends on them, what the contract says about availability, and the incidents you have seen.

How we'd frame it for leadership. AI is now a dependency like email or the ERP. Nobody expects email to never go down; they expect a plan when it does. The same standard applies here, and it is cheaper to set while the outages are short.

The 30-second version

On October 23, 2026, Anthropic's status page recorded three Claude incidents — 4:28–4:59 PM PT (critical), 5:23–5:34 PM and 10:00–10:35 PM — about 77 minutes in total across claude.ai, the Claude API, Claude Code and Claude Cowork. Claude also runs inside Microsoft 365 Copilot: Microsoft enables Anthropic models by default for most commercial customers outside the EU/EFTA and UK. We found no sign the outage reached Copilot in our own tenant, and we have not seen Microsoft say either way. Inventory where AI sits in your critical processes, write a fallback, pre-approve any alternate model, and subscribe to the status pages. If you'd like a senior engineer to map it with you, the project intake form takes about three minutes.

Sources and further reading

Pro IT NW helps regulated mid-market firms in Seattle and the Pacific Northwest govern Microsoft 365 and the AI inside it. Vendor-neutral, labor-only. Incident times are as published by Anthropic; check the status page for any later root-cause statement.

Questions we get asked

Did the Claude outage affect Microsoft 365 Copilot?: We have no evidence that it did. Microsoft's subprocessor documentation does not say how an Anthropic incident would show up in Copilot, and Anthropic's incidents listed claude.ai, the Claude API, Claude Code and Claude Cowork as the affected components. In ProIT NW's own Microsoft 365 tenant, the Service health issue history showed no Copilot incident that started on October 23 or 24, 2026. That is one tenant's view of resolved items, not a statement about every tenant — check your own Service health.
How long was Claude down?: Anthropic's status page gives three impact windows on October 23, 2026 (Pacific): 4:28–4:59 PM for Claude Fable 5.1, Sonnet 5.5 and Haiku 5.5 plus fast-mode requests to Opus 5.5, rated critical; 5:23–5:34 PM for Claude Opus 5.5; and 10:00–10:35 PM for other Claude models. That is about 77 minutes of impact across the evening, in three separate incidents.
Did the problems continue after October 23?: Yes, briefly. Anthropic's status page recorded two more short, resolved incidents over the next two days: elevated errors for Claude Sonnet 5.5 and other models on October 24 (11:05–11:35 PM PT, about 30 minutes) and for Claude Fable 5.1 on October 25 (1:08–1:29 AM PT, about 21 minutes). Both hit the same components as the 23rd. We have not seen a root cause, and we found no sign either one reached Microsoft 365 Copilot in ProIT NW's own tenant.
Is Claude switched on in our Microsoft 365 Copilot?: Possibly by default. Microsoft states it 'enables Anthropic models on by default for most customers in commercial cloud (excluding EU/EFTA and UK)', in offerings such as Microsoft Copilot, Researcher, Copilot Studio, Power Platform and Copilot in Microsoft 365 apps. Anthropic models are disabled by default for customers within the EU Data Boundary and in the UK. The setting is managed in the Microsoft 365 admin center.
Should we drop the AI vendor that had the outage?: One evening of incidents is not, on its own, a reason to switch. Every cloud service has outages — Microsoft 365 and Azure included. The useful response is to know where AI sits in your critical processes, have a fallback for those steps, and make sure any alternate model you fall back to is one your data-governance rules already allow.

Related service

Copilot readiness

Written by the team at Pro IT NW · Senior-led Microsoft project consultancy · Seattle and the Pacific Northwest, delivered USA-wide.

Share on LinkedIn Follow Pro IT NW