AI that keeps your data out of the US means the entire chain, from the language model down to the server, is neither owned by an American tech giant nor subject to American law. That may sound like a detail for the lawyers. In practice it's the first question every privacy officer asks the moment an AI tool lands on the table.
Picture a security officer at a local authority who has to approve an AI notetaking tool for the executive board's meetings. She reads the privacy page, sees "data stored in a European region" and ticks the box. A week later the data protection officer starts asking questions. The data centre really is in Europe. The parent company is in the United States. And that puts the entire service under American law, however European the server may be.
To see how that happens, it helps to take an AI service apart for a moment.
What does keeping your data out of the US actually mean?
An AI service isn't a single thing but a stack of layers. Only once you look at them separately do you see where things go wrong.
- The language modelThe model that writes the summary. Often from OpenAI, Microsoft or Google; sometimes the provider's own model.
- The speech-to-textThe step that turns the conversation into words. It runs somewhere too, and it belongs to someone.
- The cloudThe servers everything runs on, and where the data sits.
- The companyThe owner of the service, and the country it answers to. That jurisdiction decides who can legally reach your data.
Most AI tools have an American party on at least one of those layers: the model comes from OpenAI, the cloud belongs to Amazon or Microsoft, or the company itself is American. Sovereign AI means none of those layers is in American hands.
Sovereign AI, European AI, AI without an American cloud, AI without Big Tech: they're different words for the same question. Who really has your conversation in their hands, and under whose law?
What is the CLOUD Act, and does it reach EU data?
The CLOUD Act is a US law that obliges American providers to hand data over to the US government, even when that data sits in Europe. The full name is the Clarifying Lawful Overseas Use of Data Act.
That last part is the whole point. An American company falls under American law even when its servers are in Frankfurt or Amsterdam. The Dutch National Cyber Security Centre (NCSC) wrote in 2022 that the chance of the US actually invoking the CLOUD Act against European customers is small. But small isn't zero, and the provider can't rule it out (see the sources at the bottom).
In 2025 that became concrete. At a hearing of a French Senate inquiry committee, on 18 June 2025, Anton Carniaux, director of public and legal affairs at Microsoft France, was asked whether he could guarantee under oath that French customers' data would never go to the US government without consent.
"No, I cannot guarantee that, but, again, it has never happened before."
The quote is translated from the French, and it has indeed never happened, as Microsoft itself points out. But the provider can't guarantee it, and that is exactly the point: the law of the home country comes first. How to check a tool for this yourself is covered on our security page. Here we're after the why, and the why starts with a misunderstanding about the server.
Is an EU cloud or an EU data centre enough?
No, not automatically. An EU data centre determines where your data physically sits. It doesn't determine which law applies to the company that can reach it.
That difference is often glossed over. An American provider builds a data centre in Europe, sticks an EU flag and a certificate on it, and calls it sovereign. In practice that's known as sovereignty-washing: it looks European, but the parent company, and with it the ultimate say, stays American. And the odds of running into such a set-up are high: American hyperscalers hold around 70% of the European cloud market (Synergy Research Group, 2025).
"Our data sits in Europe" and "our data falls under European law" are two different promises. The first is about a place. The second is about control. For a routine meeting the difference barely matters. For a confidential conversation it's everything.
Where your conversation really goes.
The easiest way to see this is to lay two routes side by side. Not as a checklist, but as the journey your conversation makes from the moment someone presses record.
Route A: the usual route through Big Tech
- Your conversation. Someone presses record.
- To an American AI model. The tool sends the audio or the text to an American model, often OpenAI, Microsoft or Google under the bonnet.
- Stored on an American cloud. Sometimes in an EU region, but the parent company is American.
- Under the CLOUD Act. American law reaches your data, wherever it sits.
- With free consumer AI, often: training. Your input helps improve the model.
Route B: the route that stays in the EU
- Your conversation. Someone presses record.
- In-house models in Amsterdam. Speech-to-text and the summary run on our own, self-hosted models. No American party in the chain.
- The meeting report in everyone's inbox. With the summary, the decisions and the action points, and the full transcript as an attachment.
- Audio deleted after processing. Within a minute. No archive.
- No training, and your data stays in the EU. Your conversations train nothing.
The difference between the two routes isn't "better encryption". Both can be strongly secured. It comes down to who owns the chain, not to where the server happens to be. Which is why an EU flag on a website proves nothing.
When data going to the US is perfectly fine.
For a great deal of work an American AI tool is a perfectly good choice, and we're honest about that. Those tools are technically strong, well secured and connect to everything you already use. For a brainstorm, a public presentation or an internal note without sensitive content, where the model comes from is rarely a problem.
It bites in two places. Confidential conversations: a client, a patient, a case file, an executive board meeting, or a lawyer bound by confidentiality. And organisations with hard requirements around sovereignty or the GDPR. If you run conversations through a general chatbot rather than a notetaking tool, first read whether ChatGPT is GDPR compliant.
And to be fair: a European stack of your own has downsides too. Fewer integrations, no worldwide ecosystem of plugins, a smaller company behind it. This isn't a story of good against evil. It's a choice that depends on what gets said in the conversation.
How Notuly shows it can be done.
The easiest way to show it can be done is to point at something built that way. That's why Notuly is made like this: the AI notetaker for every conversation, discreet in-person, hybrid and online.
Here's what that looks like on a Tuesday. Half past two, a conversation around the table: the phone in the middle, someone presses record, and the conversation simply carries on. For meetings on Teams, Zoom or Google Meet, the desktop app for macOS and Windows does the same from your laptop. Afterwards, the report is in everyone's inbox within ten minutes: the summary, the decisions and the action points with a name and a deadline ("Sam sends the proposal to Daniël on Friday", "Leila schedules the follow-up for Tuesday"). If you choose, the transcript comes along as a .txt attachment.
What the reader doesn't see is the route underneath. The entire AI step, from speech-to-text to the summary, runs on our own, self-hosted models on servers in Amsterdam. No OpenAI, no Microsoft, no Google anywhere in the chain. The audio is deleted within a minute of processing, there is no archive, and your conversations are not used as training data. Notuly understands more than 90 spoken languages, and the report comes in Dutch or English.
On Team and above, that comes with the data processing agreement (GDPR Article 28): the document that contractually pins down the sub-processors in that chain (Article 28(2) and (4) GDPR). How we handle your data is set out in black and white under Privacy & security. That's route B, from start to finish.
For whoever has to sign this off
In short, for whoever has to approve this on privacy, GDPR and sovereignty:
- Does our data fall under American law? No. The chain sits in the Netherlands and there's no American party in it.
- Where is it processed? In Amsterdam, on our own, self-hosted models. Your data never leaves the EU.
- Is anything trained on it? No, never. Your conversations are not used as training data and are not for sale.
- Do we get a data processing agreement? Yes, available on Team and above (GDPR Article 28).
What it costs is covered in the frequently asked questions below. Still weighing the decision? Read how to choose a GDPR-compliant tool upfront.
Read how Notuly keeps notetaking secure →
Sources.
- ICTRecht (in Dutch): how the CLOUD Act works when data is stored in Europe.
- NCSC, CLOUD Act Memo and CLOUD Act Requests: the nuance that the real-world chance is small, but not zero.
- The Register: Microsoft admits it cannot guarantee data sovereignty (French Senate hearing 18 June 2025, published 25 July 2025).
- Synergy Research Group: American hyperscalers hold around 70% of the European cloud market (24 July 2025).
- European Data Protection Board (EDPB): international data transfers, on transfers to third countries under the GDPR.
Based on public sources, consulted in June 2026 and checked again in August 2026. Legislation and company policy change; check the source for the current position. This is general guidance, not legal advice.