AI for doctors: what is allowed? GDPR, confidentiality, EU AI Act and MDR

Note: this article is an overview, as of 13 October 2026, focused on the EU with examples from Germany and Austria. It is not legal advice. For concrete projects — such as introducing an AI tool in a hospital or practice — your data protection officer, legal department and medical chamber are the right contacts. Medelix itself provides an AI tool.

In short

• Patient data are health data and specially protected (Art. 9 GDPR). Pseudonymised data remain personal data.
• Professional secrecy applies to AI providers too (e.g. § 203 German Criminal Code, § 54 Austrian Physicians Act). External service providers may only be involved under strict conditions.
• Rule of thumb: do not enter identifying patient data into AI tools your institution has not explicitly approved.
• EU AI Act: the AI literacy duty has applied since February 2025, transparency duties since August 2026. Obligations for AI in medical devices were postponed to August 2028 by the "Digital Omnibus".
• AI does not replace clinical judgment. The German Medical Association calls for physicians to check AI output for plausibility.

1. Data protection: what the GDPR requires

Health data are a special category of personal data; processing them is prohibited unless an exception applies (Art. 9(1) GDPR) [1]. One exception covers medical diagnosis and the provision of care (Art. 9(2)(h)), on condition that processing is carried out by or under the responsibility of professionals bound by secrecy (Art. 9(3)) [1].

In practice:

• Pseudonymised is not anonymous. Data that can be re-attributed to a person using additional information remain personal data (Art. 4(5) and Recital 26 GDPR). Only truly anonymous information falls outside the GDPR [1].
• Processors. If an AI provider processes personal data on your behalf, a contract under Art. 28 GDPR with sufficient guarantees is required [1].
• Transfers. Processing outside the EU adds further requirements (Chapter V GDPR).

Germany's data protection conference (DSK) published guidance on "AI and data protection" in 2024, focused on large language models and intended as a checklist for choosing and using AI applications [2].

2. Professional secrecy

Germany: § 203 of the Criminal Code makes the unauthorised disclosure of patient secrets by physicians an offence. Since a 2017 reform, secrets may be disclosed to "other contributing persons" — such as external IT service providers — to the extent necessary for their work; failing to bind such persons to secrecy can itself be an offence [3].

Austria: under § 54 of the Physicians Act 1998, physicians and their assistants must keep confidential all secrets entrusted to them or learned in the course of their profession [4].

A freely available AI chatbot without a contract is generally not a "contributing person" in this sense.

3. What this means day to day

• Ask about the topic, not the patient. "Anticoagulation in atrial fibrillation with eGFR 25?" is a clinical question. A name, date of birth, case number, address or a rare combination of features turns it into patient data.
• Use only approved tools for patient data — with a processing agreement, a known processing location and approval from your data protection officer.
• Check the output. Open the cited source, check plausibility, document what your decision rests on.

As an illustration of how a provider can implement this: Medelix's terms prohibit entering patient-identifying data, and it additionally filters common direct identifiers automatically; according to its privacy policy these filters are an extra safeguard, not a guarantee [5].

4. EU AI Act: what applies when?

The AI Act, Regulation (EU) 2024/1689, applies in stages [6]. Regulation (EU) 2026/1744 ("Digital Omnibus on AI") changed several deadlines [7]. According to the summary by the Slovenian regulator AKOS [8]:

For physicians as deployers of AI systems this means:

• AI literacy (Art. 4): organisations using AI must take measures for their staff's AI literacy. The Digital Omnibus reportedly softened this duty; the consolidated text is authoritative [6][7].
• Transparency (Art. 50): providers must ensure people know they are interacting with an AI system; AI-generated content must be marked [6].
• High-risk systems (Art. 26): deployers of a high-risk AI system — such as AI that is part of a medical device — must use it according to its instructions, ensure human oversight by trained staff and monitor its operation [6]. For medical devices this applies from August 2028 under the Omnibus [8].

5. Medical device law: is my AI tool a medical device?

Software is a medical device when its manufacturer intends it for a medical purpose — such as diagnosis, prediction or treatment of disease (Art. 2(1) MDR) [9]. Software providing information used to take diagnostic or therapeutic decisions is at least Class IIa under Rule 11 [9]. Guidance MDCG 2019-11, revision 1 of which was published in June 2025, explains how software is qualified and classified [10].

What matters is the manufacturer's intended purpose. A tool offered explicitly as an information source and not as a medical device — like Medelix [11] — is meant for information and research; the clinical decision stays with you.

6. What the profession says

In January 2025 the German Medical Association adopted a statement by its Scientific Advisory Board on "Artificial intelligence in medicine" [12]. Among the ethical requirements of medical work it names checking the plausibility of AI's diagnostic suggestions [13].

Checklist before use

1. Do I need patient data for my question at all? Usually not.
2. If patient data are needed, has my institution approved the tool?
3. Where are data processed, and is there a processing agreement?
4. Does the tool cite its sources, and have I checked them?
5. Is the tool a medical device — and am I using it within its intended purpose?

More on specific tools: Medelix vs ChatGPT · OpenEvidence alternative for Europe.

Sources

1. Regulation (EU) 2016/679 (GDPR). eur-lex.europa.eu
2. Datenschutzkonferenz (6 May 2024): Orientierungshilfe Künstliche Intelligenz und Datenschutz. datenschutzkonferenz-online.de
3. § 203 StGB (German Criminal Code). gesetze-im-internet.de
4. § 54 Ärztegesetz 1998 (Austria). ris.bka.gv.at
5. Medelix: Privacy Policy. medelix.ai/privacy
6. Regulation (EU) 2024/1689 (AI Act). eur-lex.europa.eu
7. Regulation (EU) 2026/1744 (Digital Omnibus on AI). eur-lex.europa.eu
8. AKOS: Digital Omnibus on Artificial Intelligence — FAQ. akos-rs.si
9. Regulation (EU) 2017/745 (MDR). eur-lex.europa.eu
10. European Commission (17 Jun 2025): Update MDCG 2019-11 rev.1. health.ec.europa.eu
11. Medelix: legal notice. medelix.ai/impressum
12. Bundesärztekammer (2025): Stellungnahme „Künstliche Intelligenz in der Medizin“. doi.org/10.3238/arztebl.2025.Stellungnahme_KI_Medizin
13. Bundesärztekammer (21 Feb 2025): Ärztinnen und Ärzte auf Umgang mit KI in der Medizin vorbereiten (press release). bundesaerztekammer.de