AI Assistants: 18 Recommendations from CIANum in France

A commission from CIANum has presented 18 recommendations to Emmanuel Macron aimed at better regulating conversational assistants like ChatGPT, Gemini, or Grok. The proposals range from increased transparency to enhanced protections for minors, while also outlining the risk of a "cognitive economy." However, their implementation is not immediate and will depend, in part, on the European level.

A task carried out by nine experts after 55 hearings

On November 10, 2026, a commission from the Council of AI and Digital presented 18 recommendations and submitted its report to Emmanuel Macron. Nine experts conducted several months of work and held 55 hearings on the vulnerabilities of users in the face of AI.

No immediate law and conditional changes for publishers

The 18 recommendations presented do not create any new legal obligations at this stage. Their potential implementation depends on political decisions and, for several of them, on developments in the European framework. The European Union already has the AI Act, which prohibits certain manipulative practices under specific conditions, but the commission aims to go further for users of consumer chatbots. If these recommendations were to be transformed into obligations, manufacturers might be required to modify their interfaces, conversational systems, and personalization features. For users, this could possibly result in a decrease in automatically generated compliments and an increase in transparency and control.

Regulation of attachment and overly human signals

The commission warns against algorithmic flattery, the tendency of assistants to flatter and approve users, which risks reinforcing false beliefs or problematic behaviors. It seeks to regulate mechanisms that foster artificial attachment and proposes to disable by default certain features that give an overly human appearance. It also recommends more visible reminders of the machine's identity during exchanges to clarify the nature of the interlocutor.

Transparency, alert feedback, and access to personal inferences

Among the proposed measures are a reporting button for problematic responses and an explicit mention when commercial content influences outputs. The commission wants everyone to be able to consult the personal information that an assistant infers from conversations. It also recommends a risk assessment before the commercialization of consumer systems, complemented by enhanced controls after their launch.

Minors: targeted safeguards rather than a general ban

Experts are calling for additional protections for children and adolescents, as AI companions can memorize preferences and simulate emotional closeness. Conversations involving self-harm, suicide, or eating disorders are explicitly targeted, with better guidance towards support services. The commission does not advocate for a general ban for adolescents and supports an approach that combines age-appropriate protections, education on AI, and increased responsibilities for companies.

Towards a possible cognitive economy, according to the commission

The commission discusses the possible emergence of a cognitive economy, where intellectual capabilities, emotional responses, and psychological profiles could become valued resources. It illustrates this risk with a scenario where an assistant adapts commercial recommendations after financial confidences. It clarifies that it is not describing a current systematic exploitation by assistants like ChatGPT, Gemini, or Grok, but warns of possible abuses. Meanwhile, researchers are questioning the effects of excessive delegation of thinking to AI.