Can My Law Firm Use ChatGPT? §203 StGB and GDPR 2026

What §203 StGB governs — and why it applies to AI tools

§203 StGB (Violation of private secrets) makes it a criminal offence for professionals bound by confidentiality to disclose without authorisation a secret entrusted to them. Lawyers are explicitly named in the list of addressees of the provision — alongside physicians, tax advisors and auditors. The penalty range: imprisonment of up to one year or a fine. Unlike a GDPR fine, this sanction does not hit the firm as an organisation but you personally.

„Disclosure“ here does not mean only actively passing information on. It is enough to give a third party the possibility of taking note. Anyone who enters a draft pleading containing client data into a cloud AI tool whose operator can technically access the inputs creates precisely this possibility — regardless of whether any human ever reads the data.

Since the 2017 reform, §203 para. 3 StGB has expressly permitted involving „other participating persons“ — including external IT service providers. But only under one condition: the service provider must be bound to confidentiality, and the engagement must be necessary. For lawyers, §43e BRAO spells out these requirements: a written agreement, an express confidentiality obligation on the service provider, and particular care where there is a cross-border element.

What actually happens to client data in ChatGPT and Claude.ai

The free and Plus versions of ChatGPT, as well as the Claude.ai web interface, are consumer products from US companies. For use in a law firm, this raises three structural problems:

• Server location and jurisdiction: Processing takes place on infrastructure controlled by the provider — typically in the US or at least under US jurisdiction. The US CLOUD Act obliges US companies to hand over data on official order, regardless of the physical storage location.
• Training and access: In the consumer versions, inputs may — depending on the account settings — be used to train the models. In addition, providers reserve the right for staff to review conversations for abuse monitoring.
• No suitable contract: For free accounts there is no Data Processing Agreement. The business plans (e.g. ChatGPT Team/Enterprise) do offer a DPA and in part EU data hosting — but regularly no confidentiality commitment tailored to §203 StGB of the kind that §43e BRAO requires for professionals bound by confidentiality.

The three conditions for legally sound use

1. Data Processing Agreement under Art. 28 GDPR

As soon as an AI tool processes personal data, the provider is a data processor. Art. 28 GDPR requires a written contract (DPA) which, among other things, governs: processing only on documented instructions, a confidentiality obligation on personnel, technical and organisational measures, an approval requirement for sub-processors, and deletion after the contract ends. Without a DPA, every entry of client data is a data protection violation — entirely independent of §203 StGB.

2. Data processing exclusively within the EU

Both levels are decisive: the storage of your data (documents, chat histories, configuration) and the AI inference itself. An „EU data centre“ on the marketing page is not enough — ask specifically where the prompts are processed, whether data is stored for training purposes, and which sub-processors sit in the chain. The lower the US CLOUD Act exposure of the entire chain, the more robust your documentation towards clients and supervisory authorities.

3. Supplementary agreement under §203 StGB

The most common mistake: law firms review the DPA and consider the matter settled. But the DPA only covers data protection. §203 StGB additionally requires that the provider, as a „participating person“, be formally bound to confidentiality — including being put on notice of the service provider's own criminal liability in the event of a violation. Only a few providers actively offer this supplementary agreement. If your preferred provider dodges the question when asked, that is the answer.

That the question is actively occupying the professional bodies is also shown by the FAQ catalogue on AI use from the German Federal Chamber of Tax Advisors (Bundessteuerberaterkammer, January 2026): for tax advisors — who are subject to the same §203 StGB — the same review logic is laid out there. What this means in concrete terms for tax firms, we have set out in a dedicated article on ChatGPT for tax advisors.

What alternatives are there? The honest spectrum

There is no single right solution — rather three operating models with different trade-offs:

On-premise and open-source models

Maximum control: the model runs on your own hardware, no data leaves the firm. In return, you bear procurement, operation and security yourself, you need IT staff — and the freely available models usually do not reach the quality of the large commercial models for legal texts. For firms with their own IT department an option; for a five-person firm rarely realistic.

German and European SaaS providers

Providers such as Langdock, meinGPT or DeutschlandGPT operate multi-model platforms with a DPA and in part ISO 27001 certification. That solves the GDPR problem — but not automatically the §203 problem: an express confidentiality commitment under §203 StGB is not part of the standard contract for most of them. Ask explicitly before signing and have the sub-processor chain disclosed.

Managed operation: a frontier model on EU infrastructure

The third path combines the model quality of the large providers with EU data hosting and a §203 contract. This is how ClapNClaw works: Claude (Anthropic) is connected via AWS Bedrock in Frankfurt (eu-central-1) — Anthropic gets no access to your data and does not use it for training. All persistent data (documents, histories, user accounts) sits in a dedicated, isolated container at Hetzner in Frankfurt. The DPA under Art. 28 GDPR is in place from day one; the compliance tier (€59 per user/month) includes the §203 supplementary agreement as a fixed part of the contract. Which variant suits your firm depends on size, IT resources and risk appetite — a broader classification of all options can be found in our guide: using AI in a GDPR-compliant way.

Checklist: before your firm introduces an AI tool

• Immediate measure: No client data into the free web versions of ChatGPT, Claude.ai, Gemini & co. — not even „just quickly to draft something“.
• Check the DPA: Is there a Data Processing Agreement under Art. 28 GDPR? Are all sub-processors named?
• Obtain the §203 commitment: Has the provider committed in writing to confidentiality under §203 StGB (cf. §43e BRAO)?
• Document data residency: Where is data stored, where does the inference take place? Obtain written confirmation.
• Training exclusion: Have it contractually guaranteed that inputs are not used for model training.
• Internal AI policy: Define who may use which tools with which data — and train the team.
• Update the records of processing: Record the new tool under Art. 30 GDPR; involve your data protection officer, if you have one.
• Review client information: Supplement the firm's privacy notices with the AI-assisted processing.

Frequently asked questions