I've seen a lot of posts on various sites regarding this over the past few days, and most of them make the same sensible point: check your Microsoft 365 tenant before 9 December. They are right. But the default switch is only the visible half of the story, and for one group of readers it may be the less important half.
On 10 November Microsoft published a Message Center notice, MC1546435, titled "Action may be required: Claude models in Microsoft Copilot will be enabled by default in the UK". From 9 December 2026, Anthropic's Claude models will be switched on for UK tenants that have not made a choice under the admin setting "AI providers operating as Microsoft subprocessors".
The outcomes, as the notice sets them out:
| Your current Anthropic setting | What happens on 9 December |
|---|---|
| Never set | Claude enabled for all users |
| Disabled | Respected |
| All users, No users, or specific users and groups | Respected |
| Allowed only in Copilot in Microsoft 365 apps | Option retired, moves to All users |
Two lines in that table deserve more attention than they are getting.
The first is "never set". Until now, a UK tenant that did nothing had Claude switched off, because Microsoft itself excluded Claude from its EU Data Boundary and from in-country processing commitments, and treated the UK the same way as the EU. Doing nothing was a safe position. From 9 December, doing nothing becomes a decision to switch it on.
The second is the last row. An organisation that thought about this carefully, and chose to allow Claude only inside Word, Excel and PowerPoint, does not keep that choice. The narrower option is being retired, and those tenants move to all users. The careful answer gets widened without anyone in the organisation signing it off.
The notice is also plain about geography: UK tenants are not in scope for the EU Data Boundary. Microsoft has said Claude will be processed inside the EU Data Boundary by the end of 2026 for EU and EFTA customers. That promise does not cover the UK.
In September I asked a simple question in "Do You Know Where Your Copilot Comes From?": when one brand sits on top of a live mix of OpenAI, Anthropic and Microsoft's own models, can a data protection officer find out which model processed a given prompt on a given day? If not, the Article 35 assessment is describing a system nobody can observe.
To be fair to Microsoft, the user interface has moved since then. Its documentation now says that in Microsoft Copilot on web, desktop and mobile, indicators show when Claude models are in use, and in Researcher and Edit with Copilot a user can pick Claude deliberately. That is a real improvement for the person at the keyboard.
It is not the same thing as evidence. Some features use Claude by default without offering a choice, and the documentation I could find says nothing about audit records showing which model handled which request. An indicator on one user's screen is not something a DPO can produce for a regulator six months later. The September question has not been answered. It has been moved from "is Claude in the mix?" to "can you prove where it was used?"
Here is the practical consequence. Many UK organisations will have written their Copilot DPIA, their records of processing and their transfer assessments on the basis that Anthropic was not an active subprocessor in their tenant, because Microsoft's own default said so.
On 9 December, for every tenant that never touched the setting, that assumption stops being true. The documents do not change. The processing does.
Nothing in UK GDPR says personal data must stay in the UK. It does require you to know who is processing it, where, and under what safeguards, and to have assessed the risk before the processing starts, not after. A default that changes the subprocessor list on a date of the supplier's choosing is exactly the kind of change those documents exist to catch.
So the honest framing is not "Claude is a GDPR problem". It may well not be, once assessed. The problem is that on 8 December your DPIA was accurate, and on 9 December it may not be, and nobody in your organisation will have decided that.
Now the part the UK posts do not cover.
Microsoft's default-off list for Claude has always been described as "EU/EFTA and UK". Its default-on group is "most customers in commercial cloud". The Isle of Man is not in the EU, not in EFTA, and not in the United Kingdom. Microsoft's own data-location tables list it under its own country code, IM, separately from Great Britain.
I want to be clear that what follows is a question, not a finding. I have not seen an Isle of Man tenant's settings, and Microsoft's documentation does not say how it classifies Crown Dependency tenants for this purpose.
But if a tenant was registered with the Isle of Man as its country, it is worth checking whether it ever sat in the default-off group at all. If it did not, the 9 December notice changes nothing for it, because Claude may have been switched on by default since January. The same question applies in Jersey and Guernsey.
That would mean some Manx organisations have had a US-processed AI subprocessor active for most of 2026 without a decision being taken. Equally, it may turn out that Microsoft treats these tenants as UK, in which case 9 December applies to them exactly as it does across the water. Either way, the answer sits in the admin centre, and it takes five minutes to find.
Not a reversal. Three dull things.
None of this is radical. It is ordinary change control, applied to a supplier who changes its own subprocessors by default.
Enjoy your weekend.
Settings and dates are taken from Microsoft's published notice and documentation as at 12 November 2026. The Isle of Man section is a question, not a finding.
The Sovereign Auditor covers digital sovereignty, cybersecurity governance, and data protection policy, with particular focus on Isle of Man jurisdiction and Crown Dependency issues.
Payments via PayPal. Credentials delivered by email. No Substack. No Stripe. No middlemen.