INESIA (France National Institute for AI Evaluation and Security)

INESIA (Institut national pour l'evaluation et la securite de l'intelligence artificielle) is France's national institute for the evaluation and security of artificial intelligence. It was officially inaugurated on January 31, 2025, announced by Clara Chappaz, then Secretary of State for Artificial Intelligence and Digital Affairs, in the run-up to the Paris AI Action Summit held February 10-11, 2025.

Unlike many of the AI Safety Institutes established internationally, INESIA is not a standalone legal entity with its own budget or dedicated staff. Instead, it operates as a coordination and federation structure that brings together four leading French national institutions: ANSSI (the national cybersecurity agency), Inria (the national institute for digital science and technology research), LNE (the national laboratory for metrology and testing, with over 15 years of AI evaluation experience), and PEReN (the expert center for digital regulation). Joint governance is provided by SGDSN (General Secretariat for Defence and National Security, reporting to the Prime Minister) and DGE (Directorate General for Enterprise, under the Ministry of Economy).

INESIA's mission is to equip France with a robust, sovereign AI evaluation framework capable of ensuring safe innovation while protecting citizens. Its work is organized around three thematic poles and one cross-cutting axis. The Regulatory Support pole develops technical expertise for AI regulatory authorities, including evaluation methods, synthetic content detection capabilities, and AI cybersecurity certification approaches aligned with the EU AI Act. The Systemic Risks pole deepens national expertise on risks from advanced AI systems — particularly agentic systems — and contributes to France's engagement with the international AI Safety Institutes network. The Performance and Reliability pole stimulates innovation through technical challenges and cooperative evaluation exercises. A cross-cutting axis provides shared tools, technical infrastructure, and knowledge-sharing across the four member organizations.

In its first year of operation, INESIA contributed to joint work on synthetic content detection (published January 2025 through Viginum and PEReN), organized its first Scientific Days in Paris in July 2025, and participated in a joint AI agent evaluation exercise with international AI Safety Institutes partners (published July 2025). On February 12, 2026, INESIA adopted its strategic roadmap for 2026-2027, setting out its ambitions for sovereign AI evaluation capacity in France.

INESIA is part of the international AI Safety Institutes network, which includes counterparts in the United States, United Kingdom, Canada, South Korea, Japan, Kenya, Singapore, Australia, and the EU AI Office. Its unique positioning within this network stems from its mandate to operate within the EU regulatory framework (the AI Act), making it the only member institute with explicit obligations to evaluate AI systems against regulatory criteria such as reliability, robustness, transparency, human oversight, and data governance.