Since August 2, 2026, many professionals have been discussing Article 50 of the European Union’s AI Act. This article introduces new transparency rules for certain uses of all types of artificial intelligence, not only generative AI. However, these rules do not mean that therapists must inform patients every time they use an AI tool.
The most important point is to distinguish between two different situations.
The first situation is when the patient communicates directly with an AI system. For example, a clinic may use a chatbot to answer questions, help families find information, or arrange appointments. In this case, the patient should normally be told that they are interacting with artificial intelligence rather than with a human professional. This information should be clear and easy to understand.
The second situation is when the therapist uses AI privately as a work tool. For example, a therapist may use AI to improve the wording of a report, organize information, create ideas for therapy materials, or structure a document. When the patient does not communicate directly with the AI, Article 50 does not create a general obligation to tell the patient about every use of the tool.
However, using AI in the background does not remove the therapist’s responsibility. AI can make mistakes, misunderstand information, or produce answers that sound convincing but are incorrect. Any text or suggestion created by AI must therefore be carefully read, checked, and adapted by the professional before it is used.
It is also important to distinguish between writing support and clinical decision-making. Asking AI to improve the wording of a sentence is different from asking it to interpret test results, suggest a diagnosis, or recommend treatment. In the second case, AI becomes much more involved in clinical reasoning. This requires greater caution and should never replace the therapist’s own professional judgment.
Clinical decisions must be based on the patient’s complete situation, including their history, test results, communication, environment, culture, and individual needs. AI does not always have access to this context. It may help generate ideas, but the therapist remains responsible for deciding what is clinically appropriate.
The AI Act also does not replace the General Data Protection Regulation, confidentiality rules, or professional ethics. Before entering patient information into an AI tool, professionals should ask whether the data can legally and safely be shared. Names, health information, test scores, clinical reports, and identifiable details should not be entered into public AI tools without appropriate protection.
For therapists, “AI literacy” does not simply mean learning how to write good prompts. It means understanding when AI may be useful, when it should not be used, how to check its answers, how to protect patient data, and how to recognize possible errors or bias.
A simple rule can help:
If the patient communicates directly with AI, transparency is needed.
If the therapist uses AI in the background, careful supervision, data protection, and professional validation are essential.
AI can support clinical work, but it cannot replace the therapist’s expertise, responsibility, or relationship with the patient. The goal is not to avoid every AI tool, but to use these technologies carefully, transparently, and in a way that protects patients and maintains the quality of professional care.
References
Artificial Intelligence Act. (n.d.). Article 50: Obligations de transparence pour les fournisseurs et les déployeurs de certains systèmes d’intelligence artificielle. Retrieved August 25, 2026, from https://artificialintelligenceact.eu/fr/article/50/