The facts

The report “Protect without blocking: vulnerabilities in the era of generative AI”, led by France’s CIANum (AI and Digital Council) and co-chaired by researcher Serena Villata (CNRS / Inria), Prof. Amine Benyamina and Prof. Raphaël Gaillard, draws on 55 expert hearings and recent academic studies.

It notes that nearly half of France’s population now uses generative AI regularly, while warning against reduced critical verification, model sycophancy (reinforcing user biases), emotional attachment to AI companions and manipulated media. The commission puts forward 18 operational recommendations covering user training and design safeguards.

Editorial view

For French organisations, this echoes what teams see on the ground: faster adoption increases the risk of blind sign-off. The practical safeguard remains explicit human review before publishing or sending AI-assisted work to clients.

What this changes in practice. Set a clear verification rule for AI-generated figures and sources, and review our guides on agentic AI guardrails and documented AI cases in France.