“When is Claude in Copilot available in Europe?”
That’s the most frequently asked question over the last couple of months and finally there’s an answer.
Last Thursday, Microsoft announced that by the end of 2026, Claude models in Microsoft Copilot will be processed inside the EU Data Boundary, for organizations in the EU and in EFTA. Switzerland is part of EFTA.
Hey, so nice to e-meet you! Thank you for reading Copilot Your Day! My ultimate goal is that you and your organization get the most out of Copilot. Therefore I post daily on LinkedIn - if we are not connected yet 👇
Until now, that was the main reason to keep Claude turned off. Most of my customers wanted Claude in Copilot but didn’t enable it because they didn’t want their data processed in the US. In the EU, EFTA and the UK, Claude models are disabled by default and once you switch them on, the processing happens outside the EU Data Boundary. Not many of the customers I work with have switched it on. Most are waiting for the EU Data Boundary.
So there’s no final date, but the announcement say it’s “by the end of 2026”. Luckily we’re already in November and there only 7 weeks left in 2026.
The setting stays off by default, so someone in your organization still has to decide. Organizations that already use Claude move into the EU Data Boundary automatically. In-country processing for Switzerland isn’t mentioned, and neither is the UK. So here’s still no update, it was on the roadmap earlier this year, but then suddenly disappear.
So Anthrophic with its product Claude convinced so many people on the market. Not a few mentioned that Claude is the best AI tool so far. And of course many people wanted to use the greatest models in their Enterprise AI.
Claude is coming. Now what matters is who in your organization knows when to use it. That’s what this episode is about: where Claude already runs, why “Copilot or Claude or ChatGPT?” was never the right question (you heard this from me before), and which model fits which job.
Missed last week’s episode? Here’s my summary in 5 minutes (in German):
Claude in Copilot, one more time
Claude isn’t new in Copilot. Since June you can pick Claude as a model in Copilot Chat. It also runs in Researcher, in Copilot in Word, Excel and PowerPoint, in Cowork and in Copilot Code. In Copilot Studio, makers choose the model when they build an agent.
Anthropic works as a Microsoft subprocessor. According to Microsoft, that means the Microsoft Product Terms, the Data Protection Addendum and Enterprise Data Protection apply, and your data isn’t used to train the models. The exception: models labeled with data retention, like Claude Fable in Cowork. They need a separate opt-in.
The real difference between Switzerland and most other countries is the default. In most commercial tenants, Claude is on. In the EU, EFTA and the UK, it stays off until an admin switches it on. And Cowork in a tenant inside the EU Data Boundary offers GPT models only today.
That’s why many people in DACH have never seen Claude in their Copilot. And it’s behind the second question I hear every week.
Copilot or Claude? The comparison doesn’t hold
“Why do you use Copilot when Claude is so much better?”
For me, there’s a thinking error in this question. We compare models, but in daily business the platform decides. The question isn’t only which model is better. It’s also how the model gets to the knowledge you need for your work. Here’s my answer to this on LinkedIn.
And the comparison doesn’t hold for a simpler reason: Copilot is not a modell. Copilot is a multi-modell vendor, you can choose and switch while you are working. One subscription, access to the latest and greatest models from different providers.
So you don’t have to choose. I use OpenAI and Anthropic models every day. But I use them on the platform that understands my company’s knowledge.
One example. Last month I built a dynamic account app for our account team in about an hour: remaining project budgets, who is booked on which project, an overview of the running customer projects. The starting point was Copilot Cowork with Work IQ. Fabric IQ added the company data we need every day, like resource utilization and project budgets. The model behind it was the least interesting part.
The strongest model doesn’t help much if it doesn’t know the context that matters.
Which model for which job
Inside Copilot, you can still pick the model yourself or leave it to Auto. My rule for that hasn’t changed: Auto for the routine. A deliberate choice for the heavy set.
According to Microsoft, Auto weighs accuracy, speed and cost for every request and picks the model and the reasoning effort. For most everyday work, that’s exactly what you want.
Most of my day is small things, not heavy sets.
This is how I decide, case by case:
Quick questions, mail threads, meeting follow-ups, short replies: Auto. According to Microsoft, Copilot answers routine questions with a fast model and switches to a deeper reasoning model when a question needs it. Here, what Copilot finds in your emails, chats and files matters more than the model.
Long documents, complex topics, slides and analysis: Claude. Microsoft itself suggests Claude in Copilot Chat for analyzing long documents and breaking down complex topics. More on slides and analysis below.
Writing and rewriting: Auto as well.
Research across web and work data: the Researcher first, the model second. Microsoft recommends normal Chat for a quick summary or a short reply, and Researcher when you need several sources, citations and a report. In Researcher, I always use Critique: GPT drafts, Claude refines. That’s the multi-model approach in one click and I love it!
Images: nothing to choose. Cowork creates images with an OpenAI model, PowerPoint with Microsoft’s own and sometimes I switch to GPT-Image Model in Power Point directly.
Longer tasks in Cowork, when cost matters: often GPT.
When do you use Claude- and when GPT-Models? Let me know in the comments:
Two of these deserve a closer look.
Slides and analysis: Claude
In my experience, Claude is very strong in PowerPoint and in analysis tasks. The independent numbers back that up. Artificial Analysis runs a benchmark called AA-Briefcase with real deliverables: spreadsheets, presentations, documents. At the highest effort, Claude Sonnet 5.5 and Claude Opus 5.5 sit at the top, clearly ahead of GPT 6 Astra and GPT 6.1 Sol.
One catch: at medium effort, the Cowork default, Claude Sonnet 5.5 falls just behind GPT 6 Astra. So for a deck that matters, turn the effort up.
Microsoft made a similar call in its own product. The Word, Excel and PowerPoint Agents in Copilot run exclusively on Anthropic models.
Cowork: watch the credits
In Cowork, GPT models are often a bit cheaper in my experience. A test by Robert Crane from October points the same way: the same task in Cowork cost about 101 US dollars with GPT 6.1 Sol and 234 with Claude Opus 5.5. That was one run, so treat it as a signal.
The effort level costs as much as the model choice. According to Microsoft, higher effort in Cowork uses more credits. For a quick summary, the highest effort burns credits for nothing. It’s like choosing a Ferrari for a two-minute grocery run. While it can be fun, it’s also extremely expensive.
My Session from AI Productitivty Unleashed is now live on Youtube. It’s in Swiss German, so maybe anyway worth to watch it 🤓
What to decide before the end of the year
The announcement removes the strongest argument against Claude. The decision is still yours and it’s better made now than in a rush in January.
Who gets Claude? In the Microsoft 365 admin center, the setting “AI providers operating as Microsoft subprocessors” lets you enable Anthropic for all users or for specific users and Entra ID groups. If you switch it on now, the processing moves into the EU Data Boundary automatically once it’s available.
Which surfaces? Copilot in Word, Excel and PowerPoint has had its own Anthropic setting since April. For tenants in the EU, EFTA and UK created after 25 March 2026 it’s on by default, and until the change lands, that processing runs outside the EU Data Boundary. Worth a look in your tenant. Copilot Studio and Power Platform need a second allowance in the Power Platform admin center.
Which models stay off? Copilot will be your multi-model provider: Now with Claude models inside and with data retention need their own opt-in. Grok 4.7 in Cowork comes from xAI as an independent processor, which means xAI’s own data processing terms apply, not Microsoft’s.
Who tells the users? A model picker that nobody explains turns into feature tourism. People need one sentence on when Auto is enough and when they should choose. If you’re not sure, always use Auto-Mode.
My recommendation: prepare everything now and switch it on as soon as it’s available. Then push the multi-model approach in your organization and roll it out.
Remember: Access isn’t Fitness.
For a long time, the EU Data Boundary was the easy answer when someone in DACH asked about Claude in Copilot. That answer is about to expire and I think it’s great.
What stays is the harder question. Who in your organization knows when to leave it on Auto and when to choose? Remember: Access isn’t Fitness. With Claude in Copilot we get access to more models. Knowing which model fits which task is AI-fitness.
Auto for the routine.
A specific model for the heavy set.
That’s why: Let’s Copilot Your Day 🔥
PS: Want to try Cowork hands-on with your sales team? We are running “Cowork in a Day”, an onsite workshop in Zurich for sales professionals, on 30 November and 14 December: get here all the information
Take it with you
Cheatsheet: The Copilot Toolbox: when to use Chat, an app or Cowork, on one page.
Cheatsheet: Chat or Cowork?: the decision before the model decision, as a one-pager to download.
Cheatsheet Model or Auto: How to decide when to use Auto-Mode, or a specific model.
Where to Go Deeper
Guide: When to Use What: how to pick the right Copilot tool for the job.
Episode 119 — Microsoft-bashing is cool right now: why I still bet on the platform.
Microsoft Copilot Blog: Bringing Claude models in Microsoft Copilot into the EU Data Boundary: the announcement from 5 November.
From the archive
Too Long; didn’t read & FAQ
AI-generated summary and Frequently asked questions
TL;DR
By the end of 2026, Claude models in Microsoft Copilot will be processed inside the EU Data Boundary, for the EU and EFTA, including Switzerland. That removes the main reason to keep Claude switched off in DACH. It also shows why “Copilot or Claude?” was never the right question: Claude already runs inside Copilot. The real questions are who gets it, and who knows which model fits which task.
The basics:
Microsoft announced it on 5 November. No exact date yet, and the setting stays off by default.
Today, Claude is off by default in the EU, EFTA and the UK, and its processing runs outside the EU Data Boundary. Cowork inside the EU Data Boundary offers GPT models only.
Claude in Word, Excel and PowerPoint has its own setting, on by default for EU, EFTA and UK tenants created after 25 March 2026.
The strategic implications:
Auto for the routine. A deliberate choice for the heavy set.
Claude leads independent benchmarks for slides, spreadsheets and documents. Current GPT models often use fewer credits in Cowork.
In Cowork, the effort level matters as much as the model. Higher effort means more credits.
Prepare now (who gets Claude, on which surfaces, which models stay off, who explains it), switch on as soon as it’s available, then roll out a multi-model approach.
Is Claude in Microsoft 365 Copilot covered by the EU Data Boundary?
Not yet. Today, Claude models in Copilot are excluded from the EU Data Boundary and turned off by default for tenants in the EU, EFTA and the UK. On 5 November 2026, Microsoft announced that Claude models in Copilot will be processed inside the EU Data Boundary by the end of 2026, for customers in the EU and EFTA, which includes Switzerland. Until then, organizations inside the EU Data Boundary can only use GPT models in Copilot Cowork.
How do admins enable Claude in Microsoft 365 Copilot?
In the Microsoft 365 admin center, go to Copilot, Settings, View all, and open “AI providers operating as Microsoft subprocessors”. There you can enable Anthropic for all users or for selected users and Entra ID groups. Copilot Studio and Power Platform need a second allowance in the Power Platform admin center. Claude in Word, Excel and PowerPoint has its own setting, which is on by default for some newer tenants.
Which model should I use in Copilot Cowork?
Start with Auto. According to Microsoft, it weighs accuracy, speed and cost for every request and picks the model and the reasoning effort. Choose deliberately for heavy work: Claude models lead independent benchmarks for deliverables like presentations and spreadsheets, while current GPT models such as GPT 6.1 Sol often use fewer credits in Cowork. Pascal Brunner-Nikolla’s rule: Auto for the routine. A deliberate choice for the heavy set.
What is Critique in Researcher?
Critique is a Researcher mode that combines models from two providers: GPT drafts the response, and Claude reviews it for accuracy, completeness and sources before you see it. Microsoft also offers Council, which runs one prompt across several models in parallel and shows where they agree and where they differ. Pascal Brunner-Nikolla uses Critique for his Researcher work.
Does the model choice affect Copilot Credits?
Yes. Cowork consumes Copilot Credits, and according to Microsoft, higher reasoning effort is more thorough but slower and uses more credits. Usage also depends on context volume, orchestration and the tools used. Outside Cowork, the Microsoft 365 Copilot license includes some models with fair use, like GPT 5.6 and Sonnet, and others with limits, like Opus, while frontier models such as Astra and Fable consume Copilot Credits.
Are all models in Copilot covered by Microsoft’s data protection commitments?
No. Anthropic operates as a Microsoft subprocessor, so the Product Terms, the Data Protection Addendum and Enterprise Data Protection apply, with an exception for models marked as data retention, such as Claude Fable 5.1. Grok 4.7 is provided by xAI as an independent processor under its own data processing commitments. Admins can turn off individual models in Cowork.




