What happens to your data when we work on it
Before you let us near client data, you want to know where it goes, who can see it and what happens when the project ends. That's below.
We sign first, then we get access
Before we touch client data, we sign a data processing agreement and, if you want one, a non-disclosure agreement.
We sign both before we start. The data processing agreement says which data we process and why, where it's stored, for how long, which other services are involved (sub-processors) and what happens to the data at the end. If you have your own template you'd rather use, we read it and sign that.
Your data stays in the EU, on your accounts
This is what a typical setup we build looks like.
What we build runs on your own cloud account, in a European region by default, such as AWS in Frankfurt (eu-central-1). We only pick a region outside the EU if you ask for it in writing, and then we tell you what that means under the GDPR.
For each project you get the list of sub-processors: every service that sees your data, with the region it runs in.
How we choose an AI service
We tell you beforehand which service we use, and in which region. It has to meet three conditions.
- Processing in the EU. A European endpoint or an EU region, with no routing to regions outside it.
- No training on your data, in the contract. It's in the service's terms, so it doesn't depend on a setting someone could switch later.
- A data processing agreement from the provider. So the service can go on your sub-processor list.
Two examples of services that can do this when set up correctly. Microsoft states that for models sold by Azure, prompts and completions are not available to OpenAI and are not used to train models. With a "DataZone" deployment in an EU member state, processing stays within the EU. For Amazon Bedrock, AWS states that model providers have no access to your prompts and completions.
One caveat: some models require the provider to keep your prompts for a while to detect abuse. On Azure, Microsoft may store prompts for human review when its system suspects abuse, unless you apply for and get modified abuse monitoring. On Bedrock, AWS keeps them for up to 30 days for certain models. If that applies to the model we propose, it says so in the proposal, with an alternative that keeps nothing.
Sources: Microsoft, Data, privacy, and security for Foundry Models sold by Azure; AWS, Amazon Bedrock data protection and data retention. Checked on 25 October 2026. Provider terms change; we check them again for every project.
Health data only after an explicit decision
Health data, and the other special categories in the GDPR (such as religion, trade union membership or biometrics), don't go to an AI model unless you decide so explicitly and in writing. You make that call after we've looked together at what happens and what can go wrong.
If health data goes to an AI model, a data protection impact assessment (DPIA) is usually needed: sensitive data plus new technology. We draw it up with you, before anything gets built.
Source: GDPR, Article 9 (special categories), Article 35(1) (DPIA) and Recital 91.
Code, accounts and data are in your name
That's in the quote and in the contract. The code lives in a repository owned by your organisation. The cloud account and the keys for AI services are in your name and on your billing. We get access as users, and you can revoke that access at any time.
If you stop working with us, nothing that runs changes. How it works is written down, so another developer can take over.
Our access is per project and ends with it
- Whoever works on it for us gets their own account with two-factor authentication, with no shared passwords, and we keep no copy of your data.
- We only ask for the permissions the work needs, for this one project.
- Where we can, we test with made-up or anonymised data.
- At the end we remove our accounts and keys. You get a list of what we removed.
If something does go wrong
If we notice a possible data breach, we tell you without undue delay, as the GDPR requires of a processor. You then report it to your data protection authority within 72 hours of becoming aware of it, unless the breach is unlikely to result in a risk to the people concerned. We help you with that report.
Source: GDPR, Article 33.
NIS2 probably doesn't apply to you, but it does to your customers
Most small companies aren't directly in scope of NIS2. The directive applies to organisations in the sectors listed in Annexes I and II, starting at medium-sized enterprises: 50 staff, or both annual turnover and balance sheet total above €10 million. There are exceptions where size doesn't count, such as DNS service providers and trust service providers.
Your customers who are in scope have to look at the security of their suppliers. That's why their questionnaire lands on your desk. We help you fill it in for everything we build and run, and point out which questions are about your own organisation.
Source: Directive (EU) 2022/2555 (NIS2), Article 2 (scope) and Article 21(2)(d) (supply chain security).
What we don't do
- Paste client data into a free or personal account of an AI service.
- Use your data for anything other than your project, or to train a model.
- Add an AI service or sub-processor without telling you first.
- Build a tool that makes decisions about your customers or staff on its own. A person always checks.
- Stamp "GDPR-proof" on anything. We say what we checked and what we didn't.
- Keep access after the project is over.