Artificial intelligence (AI) is widely used in the daily lives of researchers. The Assistance publique-hôpitaux de Paris (AP-HP) therefore offers recommendations on the proper use of AI systems in medical research for the writing of scientific articles.
How to regulate the use of AI for scientific writing? At the Assistance publique-hôpitaux de Paris (AP-HP), the scientific integrity office, set up in 2022, has drafted a charter for the responsible use of generative artificial intelligence (AI) (a technology that creates new content, articles, images, sounds, from large quantities of data) in the writing of scientific articles. This note aims to “ supervise the use of generative AI tools and recall the principles that it is essential to respect for responsible use“, specifies the AP-HP. “ They apply to the preparation of manuscripts and will help preserve, during this important stage, the ethics, scientific integrity, transparency, confidentiality and reliability of your scientific work.« .
According to this document, generative AI can speed up writing and translation, but it should never replace the human judgment, ethics and responsibility of authors. He thus recalls that “ the human operator (the researcher) is always the only one responsible for the final result » and must never “use this tool to make vital decisions for them”.
Accountability, data security and scientific integrity
Several aspects are explored, starting with that of the responsibility of the authors. “ AI is a tool: your responsibility therefore remains fully committed to what you produce, whatever the role that AI has played. This obvious principle is binding: you will have to ensure yourself that the AI has not manufactured, falsified or plagiarized the content it has offered you. This means that you must, more than ever, maintain a critical approach to what you have developed with its support. Check the accuracy of the references used, – and their reliability. The same goes for quotes. Additionally, AI should never replace your scientific analysis or judgment. Its role must remain strictly auxiliary: it can facilitate the structuring or reformulation of certain passages, but never interpret results for you« .
The AP-HP also highlights the importance of data security. “ Never transmit nominative or pseudonymized patient data, as well as content relating to intellectual property not yet protected. Generative AI models use user data to respond and progress, so care must be taken not to provide any data that could lead to a breach of medical confidentiality“, specifies the text in particular.
Finally, this note raises the question of scientific integrity. “ As with all tools, you will need to be as transparent as possible to describe in the manuscript how you used AI and to what extent (translation, statistics, extensive writing, simple proofreading, etc.) Do not use generative AI in a substantial way in sensitive activities that could have an impact on other researchers or organizations« .
À retenir
Si l'IA générative est une puissante machine à accélérer la science, son utilisation dans la rédaction médicale scientifique doit être perçue comme la manipulation d'un instrument chirurgical : l'opérateur humain (le chercheur) est toujours le seul responsable du résultat final, doit s'assurer que l'instrument est stérile (confidentialité) et n'utilisera jamais cet outil pour prendre des décisions vitales à sa place (évaluation par les pairs).
To find out more:
•Note on the use of generative AI for writing scientific articles, AP-HP, May 2026.
