While nearly one in two French people now spontaneously turn to artificial intelligence for information, health issues remain fragile ground. According to the 2026 Digital Barometer cited by France Assos Santé, 48% of French people already use AI as a reflex to search for information, while the HSL19 analysis from Santé Publique France shows that 46% of adults consider it difficult to find and use health information online. In this context of hesitant confidence but massive use, a leading institution has decided to establish a clear framework for the general public around AI in health.
On June 18, 2026, the High Health Authority published, with the CNIL and France Assos Santé, new benchmarks dedicated to users of theGenerative AI in health. At the heart of this system, a printable flyer entitled
Artificial intelligence in health: use it well and protect yourself well and a FAQ in seven entries explaining how these tools work, their possible contributions and their limits. These texts extend a first document published in the fall of 2025 for caregivers and mark a step: making concrete advice available to everyone for using AI without putting themselves in danger.
Why the HAS is warning about generative AI in health
The High Authority for Health describes generative AI as a tool capable of producing texts, images or medical summaries based on probabilities learned on immense volumes of data. Their answers are not guaranteed to be true: they correspond to what the system judges most likely and may include errors, biases or invented sources.
For HAS, two risks dominate. On the one hand, reliability: if the AI relies on false, old or unverified information, the responses can become partial, erroneous, or even dangerous for health. On the other, the personal data : many general public tools keep questions and answers, sometimes outside the scope of the
GDPRwhich complicates the exercise of users’ rights.
Practical HAS, CNIL and France Assos Santé guidelines for users
The HAS flyer and FAQ detail concrete situations: using AI to better understand a report, deciphering a blood test while taking care to hide name, date of birth or contact details, or even preparing for an appointment. The institution reminds that only a medical professional can establish a diagnosis and prescribe tests or treatments.
The benchmarks emphasize a few reflexes: formulate precise questions, systematically ask the tool to cite its sources and consult them, keep in mind that the answer is not a search engine result but a reformulation. They also recommend finding a balance between sharing useful medical information and protecting one’s identity, for oneself as well as for one’s loved ones or caregivers.
Generative AI in health: a tool for discussion with the doctor
HAS also points out that professionals themselves use AI tools, often more specialized than those of the general public. When a caregiver uses a medical device incorporating AI for prevention, diagnosis or care, article L.4001-3 of the Public Health Code requires them to inform the patient.