A commission of experts delivered Emmanuel Macron a report containing 18 proposals to better monitor AI chatbots (ChatGPT, Claude, Gemini, Grok, etc.). Protection of minors, prevention of dependency, commercial transparency and model control: the recommendations aim to reinforce the obligations of the companies that develop these services.
This commission was formed by the Council for Artificial Intelligence and Digital Technology (CIANum) at the request of the government. Led by nine experts from different disciplines, it interviewed in particular 55 health professionals, researchers, companies and child protection associations. Their report identifies several risks associated with the increasing use of conversational assistants.
Chatbots with less flattering AI and more protections for minors
The commission is particularly concerned about mechanisms that can create an emotional attachment to artificial intelligence services. In particular, it recommends banning design techniques that exploit psychological vulnerabilities to encourage compulsive use. Therefore, features that simulate human emotions, excessive flattery, or give the impression of an exclusive relationship should be disabled by default.
For minors, the proposals go further: creation of a dedicated mode, limitation of usage time, ability to immediately interrupt a conversation and deactivation of persistent conversational memory. Supporters should also better detect conversations that mention suicide, self-harm or eating disorders, to direct users to appropriate support services.
Controls before and after marketing
Another important proposal: impose an independent evaluation of AI chatbots before their availability. Companies like OpenAI and Anthropic should demonstrate that their models do not exploit user vulnerabilities. Periodic checks would then be necessary to check its behavior after its launch, especially during important changes to the models.
The commission also wants to give authorities and authorized researchers better access to usage, training and monetization data. Users should be able to report problematic responses through a common button, while providers would be required to better document how their services work.
The commercial issue is not forgotten. The report calls for a ban on hidden advertising in attendees’ responses and clear labeling of commercially influenced content. It also asks companies to make accessible the information that their AI deduces from conversations, particularly about users’ health, political opinions or financial situation. Everyone should be able to consult these inferences, question them and reject their commercial exploitation.
It’s not about banning ChatGPT, Claude or Gemini
The report does not advocate a complete ban on chatbots with artificial intelligence, unlike the debates surrounding social networks among those under 15 years of age. The CIANum considers that these tools are also of interest, particularly in the educational field. On the other hand, it recommends strengthening education in artificial intelligence from primary school and developing models specifically adapted to students.
Finally, experts want the State to equip itself with the means to quickly detect disinformation campaigns generated by AI and prepare a code of good conduct for candidates and political parties ahead of the 2027 presidential elections.
These 18 recommendations remain proposed. Its implementation will require legislative and regulatory developments with European coordination to reinforce the obligations imposed on public AI services in general.
