Public research body · Science policy · Est. 2025
National Institute for AI Evaluation and Security (INESIA, Institut national pour l’évaluation et la sécurité de l’IA) is a French public “national anchor” created in 2025 to build sovereign, public capacity for evaluating advanced AI systems and for supporting AI security and trust in regulatory and operational contexts. INESIA is jointly piloted by France’s Directorate-General for Enterprises (DGE) and the General Secretariat for Defense and National Security (SGDSN), and it federates expertise from ANSSI, Inria, LNE, and PEReN to develop evaluation methods, metrics, and technical infrastructure, while also coordinating work that positions France within international networks for AI evaluation.
Strategically, INESIA’s 2026–2027 roadmap organizes activity around three program pillars—support for regulation; systemic risks; and performance & reliability—plus a cross-cutting workstream focused on creating shared “commons” (knowledge bases, technical architecture, and scientific events) and ensuring access to an AI evaluation framework and infrastructure for reproducible testing (including for sensitive models).
Publications are not yet indexed for this organization. Read them on inria.fr ↗.
Inria describes INESIA’s roadmap and its three thematic areas (support for regulation, systemic risks, performance & reliability) and cross-cutting focus, framing INESIA as a pillar of France’s national AI strategy.
INESIA adopts its roadmap for 2026–2027SGDSN and Inria report that INESIA has reached a milestone by adopting strategic priorities for 2026–2027 to provide France a robust sovereign AI evaluation framework.
SGDSN: INESIA — Evaluer et sécuriser l’IA (overview of mission and governance)SGDSN provides a consolidated description of INESIA’s creation date, joint piloting (SGDSN and DGE), its three strategic poles, federated ecosystem partners (ANSSI, Inria, LNE, PEReN), and its role in representing France within the international network of AI Safety Institutes.
Inria interview: on the national AI strategy and INESIA’s role (including regulatory/A.I. Act context)Inria’s interview discusses INESIA’s regulatory-operating context (AI Act/RIA) and mentions contributions such as work on cybersecurity risk reporting and a leaderboard effort referenced alongside national AI coordination and partner organizations.
Economie.gouv.fr: Leaflet/feuille de route for INESIA (2026–2027)Economie.gouv.fr reiterates the adoption of INESIA’s 2026–2027 roadmap and the institute’s mission to build sovereign AI evaluation capacity across the relevant ecosystem partners.
SGDSN PDF roadmap: INESIA roadmap 2026–2027The roadmap PDF provides detailed project descriptions, including evaluation infrastructure access (Project N°10), cross-cutting knowledge pooling (Project N°9), and scientific events for advancing the science of evaluation (Project N°11).