AI in practice and hospital: patient data, medical confidentiality and the vendor question

Which rules apply to patient data and AI?

• Art. 9 GDPR: Health data falls into the special categories of personal data — processing is permitted only under narrow conditions, and the requirements for security and vendor selection are correspondingly high.
• § 203 StGB / medical confidentiality: Doctors, dentists, psychotherapists and pharmacists are professionals bound by secrecy. Passing patient information to a service provider is permitted only where it is necessary and the provider has been bound to confidentiality (2017 reform; supported in professional law by the medical codes of conduct, cf. § 9 MBO-Ä).
• Processing on behalf & third countries: A DPA under Art. 28 GDPR is mandatory; with US providers the third-country problem is added (Art. 44 et seq., Schrems II, CLOUD Act). The German Data Protection Conference set out its expectations for AI systems in guidance published in 2024.
• Distinction from medical devices: AI that supports diagnostic or therapeutic decisions falls under medical device law (MDR) and the high-risk rules of the EU AI Act. Administrative assistance — text, summaries, organisation — does not.

Why are public chatbots risky in practice?

A discharge letter in a public chatbot is a disclosure of patient secrets to a provider that stores the input, may use it for training and processes it in the USA. With health data this is doubly critical: under criminal law (§ 203 StGB) and under data protection law (Art. 9 GDPR — where supervisory authorities are strictest). Here, too, the practical risk is called shadow AI: team members use private chatbot accounts because there is no official, compliant tool.

What can a practice or hospital actually use AI for?

1. Documentation: turning bullet points into structured drafts of discharge letters, findings reports and progress notes — the biggest time saving in day-to-day practice.
2. Patient communication: information and consent texts in plain language, appointment reminders, draft replies to recurring enquiries, translations for patients who speak other languages.
3. Summaries: condensing extensive prior findings, discharge letters or guideline updates to what is relevant.
4. Organisation: quality management documents, team meeting minutes, work instructions, training materials.
5. Billing & administration: draft texts for appeals, letters accompanying expert opinions, correspondence with payers.

Not included: diagnosis, treatment recommendations, medication decisions. A chat assistant is neither intended nor approved for those.

How PrivatAI meets the requirements

PrivatAI stores nothing: inputs and responses are processed and discarded — not logged, never used for training; no body of data arises that could be affected in an incident. Processing exclusively within the EU: the application in Germany (Hetzner), the AI processing in France (Scaleway). Contractually secured through the DPA under Art. 28 GDPR with a professional-secrecy clause for the healing professions (§ 203 StGB) and, on request, a separate undertaking on professional secrecy. For organisations with elevated requirements (e.g. hospitals) there is an Enterprise setup with a contractual § 203 commitment across the entire processing chain and German inference hosting. Integration into practice software via the OpenAI-compatible API, everyday work via the chat.

Which plan covers what. Essential and Professional include the EU processing described above and the Art. 28 GDPR data processing agreement with its professional-secrecy clause — that clause binds PrivatAI. What the standard plans do not include is the § 203 obligation passed down across the entire sub-processor chain, AI inference in Germany, and assurances for seizure situations (§ 97 StPO). Those are part of the Enterprise setup only and are agreed separately. For a practice this is usually not the deciding factor; for a hospital with its own risk assessment covering the full processing chain, it is.

PrivatAI supplies the technical and contractual prerequisites for a compliant setup. Whether a particular use is permissible under professional law in your specific case is a question for your Ärztekammer or your own data protection officer — no provider can certify that on your behalf.

PrivatAI (privatai.com) — AI chat & API from Germany. No content logs, no training on your data, GDPR-compliant.

PrivatAI compared (default settings)