France's AI governance faces prompting pitfalls and new Codema standards

Generative AI deployments in French businesses are hampered by poor prompting, which multiplies queries, raises costs, and increases carbon emissions. Experts stress that teams must first verify AI suitability, select appropriate models, and craft clear, structured prompts that define role, context, audience and objectives to achieve reliable, reproducible results.

In parallel, the standards organization Codema has launched two technical committees (TC23 and TC24) to create continuous‑evaluation and audit methodologies for AI systems. The initiative responds to a survey showing only 30 % of firms have advanced AI governance and aligns with the EU AI Act and France’s “Osez l’IA” plan. Codema aims to produce matrices, protocols and an “audit‑as‑a‑layer” framework that continuously collects verifiable evidence throughout an AI system’s lifecycle, helping companies demonstrate transparency, human oversight and risk management.