Local AI for confidentiality professions (§203 StGB)

Timo Wevelsiep•Updated: 23.07.2026

Editorial note: Versions, commands and prices may change. Please verify critical steps independently before production use. This guide does not replace individual consulting.

Have AI set up in your firm or practice, GDPR- and secrecy-compliant? WZ-IT builds and operates local AI systems where client and patient data never leaves the building - from one team, GDPR-compliant. See managed AI

For most companies, AI data protection is a trade-off. For confidentiality professions it is a criminally sanctioned duty: where attorney-client privilege, medical confidentiality or tax secrecy apply, dealing with AI is not optional. This article explains why cloud AI is not an option here, how local AI solves the problem and what a clean setup looks like. As of July 2026.

What makes confidentiality professions special
Why cloud AI is not an option here
Local AI as the compliance solution
The professions at a glance
What a clean setup looks like

What makes confidentiality professions special

Certain professions are subject under §203 StGB (the German Criminal Code) to a criminally sanctioned duty of secrecy that goes beyond general data protection obligations. These include, among others, lawyers, doctors and dentists, pharmacists, psychotherapists, tax advisors, auditors and notaries - as well as their professional assistants.

For them the unauthorized disclosure of protected information is not only a data protection breach but a criminal offense. That fundamentally changes the starting point for deploying AI: the standard is not "is this defensible under data protection law" but "is it ruled out that protected information reaches unauthorized parties".

Why cloud AI is not an option here

Cloud AI processes inputs on the servers of an external provider, often in a third country. For confidentiality professions this is doubly problematic: it can constitute an unauthorized disclosure within the meaning of §203 StGB and at the same time a GDPR breach, because protected data is transmitted without a sound basis.

A common fallacy is that a data center in Germany solves this. What is decisive, however, is not the location but the control: a US provider is subject to US law, even with a server in Frankfurt (see AI sovereignty). Where client, patient or tax secrecy applies, cloud AI is therefore usually not an option.

Local AI as the compliance solution

The solution is structurally simple: the data does not leave the building. An open language model runs on your own or controlled infrastructure, requests and documents are processed locally, and nothing goes to an external AI provider (see What is local AI?).

This preserves professional secrecy - patient data never leaves the practice, case files never leave the firm. The move away from cloud AI can also be justified and documented cleanly in the record of processing activities. For knowledge systems a further requirement is added: permissions must be enforced technically, so that an employee only sees via the AI what they may otherwise see - the path is in RAG with real permissions.

The professions at a glance

The requirements are similar, the details differ by profession - from professional codes to sector-specific rules. For the individual professions we have prepared the AI use concretely:

In healthcare the telematics infrastructure additionally governs how patient data may be handled, in the financial sector BaFin and the EBA require close control over outsourcing - more reasons to operate AI under your own authority.

What a clean setup looks like

A dependable AI setup for confidentiality professions consists of a few clear building blocks:

A local model on controlled hardware - GPU server, Proxmox or bare metal, nothing in someone else's cloud.
An interface with rights management - users, roles and groups, connected to your own directory.
Technically enforced permissions for knowledge systems - not left to the model via a prompt.
Logging for evidence, for example via an observability layer.
Documentation in the record of processing activities.

Operation and maintenance of this stack can be outsourced as a managed service without giving up data authority - which is exactly the difference between "AI forbidden, too risky" and "AI made usable within professional secrecy".

Rather have it operated?

You'd rather not run Local & Sovereign AI yourself? WZ-IT handles setup, operations and maintenance - GDPR-compliant from Germany.

Ai Server →

Frequently Asked Questions

Answers to the most important questions

Yes, but not any AI. Lawyers, doctors, tax advisors, psychotherapists and other professions under §203 StGB are subject to strict professional secrecy. AI is permitted as long as the protected information is not disclosed without authorization. That is precisely the problem with cloud AI - with local AI that keeps the data in-house, the use can instead be arranged cleanly.

Because the protected data is transmitted to an external provider - often to a third country. For confidentiality professions this can constitute an unauthorized disclosure within the meaning of §203 StGB and at the same time a GDPR breach. Where client, patient or tax secrets are affected, cloud AI is therefore usually not an option.

By keeping the data in-house. An open language model runs on your own or controlled infrastructure, requests and documents are processed locally, nothing goes to an AI provider. This preserves professional secrecy, and the use can be documented cleanly in the record of processing activities.

Among others lawyers, doctors and dentists, pharmacists, psychotherapists, tax advisors, auditors and notaries, as well as their professional assistants. They are subject to a criminally sanctioned duty of secrecy that goes beyond general data protection obligations - and that demands particular care in deploying AI.

Not automatically. What is decisive is not the server location alone but who controls the provider and infrastructure. A US provider with a data center in Germany remains subject to US law. Sovereignty - and thus the dependable basis for secrecy compliance - arises only when provider and infrastructure are under your own or European control.

A local language model on controlled hardware, an interface with user and rights management, for knowledge systems a retrieval service that enforces permissions technically, and logging for evidence. Plus documentation in the record of processing activities. Operation and maintenance can be outsourced as a managed service without giving up data authority.