How we protect your data
Security is the first thing we think about when we build a system. Here we describe how we work with your data and which parts of the security setup we are still building.
How we work
We break the process into small steps
We break every process down into small steps. A small step is easier to check, and we catch an error where it occurred.
AI only where it is truly needed
Wherever possible we write plain, predictable code that gives the same result every time with the same data. We use an AI model only in the steps that cannot be done any other way. This reduces the risk of errors, costs and unpredictability.
External models get de-identified data only
For complex tasks that need more powerful AI models, we de-identify the information first. After processing, it is saved in a local database and deleted from the external providers’ servers.
Your data stays yours
We do not train any AI models on your data and do not show it to other clients, not even as an example.
Where your system runs
For sensitive data we offer two options. In both cases you rent the server from us, so you do not need to buy one. We agree on which to choose during the call, before we see any of your documents.
Encrypted server in the EU
Data is processed on servers located in the EU and is not transferred outside the EU.
Server at your premises
The server stands at your premises, so all data is kept with you. The system’s internet access is strictly limited; the data is encrypted and belongs only to you.
| Encrypted server in the EU | Server at your premises | |
|---|---|---|
| Where the server is | In a data centre in the EU | At your premises |
| Where the data is kept | On servers in the EU | With you, encrypted |
| How you pay | You rent it from us | You rent it from us |
What we are still building
To be open about it: we have not finished the whole security infrastructure yet. It is a complex project that we started recently, so we are building it in parts.
We plan to create a separate “vault” for each client, in which their data will be encrypted in three layers. One client’s data will not be kept together with another’s.
In the future we plan to host powerful open-source AI models on our own servers, so that even the most complex tasks do not leave your private environment.
You can ask how far the work has progressed during the call. When the vault is running, we will say so on this page.
When the data is not highly sensitive
In that case we can deploy external tools and services. We choose only business plans whose terms forbid the provider from training models on your data. Deployment is faster and cheaper. The data is processed on the provider’s servers under that provider’s data processing and security terms.
What we do with your documents
During the audit we do not collect sensitive data; we only map the processes and the workflow. The audit findings can be deleted at your request.
Where we keep them
Documents and sensitive data are kept on your servers or in encrypted storage.
Who can open them
Your documents and sensitive data are kept separately from the system itself. We maintain the system, but we have no direct access to your data. The encryption key can be kept with you. A backup copy of the key may be kept with us, but both the key and the copy are protected by your password and strong encryption. We cannot open your data without you.
What we don’t do
We do not train any model on your data and do not show it to other clients. We do not upload it to tools whose terms allow the provider to read it.
Deletion
If you want us to delete your data, or want to know what data of yours we hold, write to [email protected].
Non-disclosure agreement
Before the first working meeting, MB „Sv projects“ and your company sign a non-disclosure agreement. This page describes our principles for working with data, while specific obligations are defined by the signed agreement.
Questions about security?
In a free 30-minute call we will discuss what data you handle and which option suits you. You do not need any documents for this.
You can also write to [email protected] or call +370 630 33090.