France - National AI Strategy

Overview

The Villani Report ("Donner un sens à l'intelligence artificielle"), commissioned by the Prime Minister and delivered on 28 March 2018, was the output of a parliamentary mission run from 8 September 2017 to 8 March 2018. The mission analysed the scientific, economic, social and ethical dimensions of AI and produced 235 pages of findings and recommendations intended to shape a national and European strategy. The report proposes a state-led combination of research investment, data policy, sectoral initiatives and regulatory/ethical frameworks. It served as a blueprint for public announcements at the "AI for Humanity" summit and informed France's national AI commitments. The full official text is available at Villani Report (PDF) on Vie-publique and related summaries are published by the Prime Minister's resources and ministerial sites.

Definitions

For the purposes of the report, "artificial intelligence" refers broadly to contemporary statistical learning and algorithmic systems used for decision support, prediction, recognition and automation across public and private sectors. The report distinguishes between research (theoretical and algorithmic advances), infrastructure (compute and data storage), applications (sectoral deployments) and human factors (labor impacts, skills, ethics). It uses operational terms such as "data policy" (rules and practical measures for sharing, access and governance of datasets), "3IA" (interdisciplinary AI institutes), "sandbox" (regulated experimental environments) and "discrimination impact assessment" (a proposed audit-style instrument to test for biased outcomes).

Governance and Institutional Framework

The Villani Report recommends a multi-level governance architecture combining national coordination, cross-ministerial leadership and specialized institutional actors. Key governance recommendations include appointing an interministerial coordinator to implement the strategy; creating a shared AI competence hub within the state to offer common tools and services; and establishing a network of 4–6 interdisciplinary AI institutes (3IA) to federate research, education and industry collaboration. The report calls for dedicated governance for sensitive sectors (health, defence) and stronger interfaces between research actors and private industry. It expressly recommends leveraging public procurement as a tool of industrial policy and suggests institutional arrangements for data-sharing, including sectoral platforms and regulated sandboxes. For context and access to the primary institutional recommendations see the official publication at Villani Report summary on Vie-publique.

Key Focus Areas

The report concentrates effort on a small number of high-value priorities: 1) a data-centric economic policy (encourage mutualisation of data, targeted opening of private datasets, portability for citizens), 2) research capacity building (3IA network, supercomputing resources and research cloud access), 3) sectoral acceleration in health, transport, environment and defence, 4) workforce and education (tripling AI-trained personnel and fostering diversity), and 5) ethics and inclusion (explainability, audits and public participation). It stresses that data availability and compute power are central levers and proposes instruments such as labels and guichets (single information windows) to consolidate domestic AI supply. In health, the report recommends platforms to enable safe experimentation with clinical data; in transport it proposes European-level innovation platforms for autonomous vehicles; in environment it proposes data infrastructures to measure digital environmental footprints; and for defence it recommends sovereign capacities and experimentation frameworks. Across sectors, it emphasizes interoperability, standards and the need for sector-specific data governance mechanisms. The report also sets out concrete recommendations for public procurement to catalyse domestic AI firms and proposes legal/contractual mechanisms to mitigate vendor lock-in and protect public interest when the state is a buyer of AI-driven systems.

Implementation Framework

Implementation in the Villani Report is framed as a set of staged actions: initiate (call for projects to create 3IA hubs); invest (supercomputing and research funding); deploy (sectoral sandboxes and experimental platforms); and regulate (ethics, audit and transparency mechanisms). The report recommends a mix of public funding commitments, public–private partnerships (including industrial chairs and co-funded projects), regulatory adjustments to enable controlled sharing of data, and the use of state purchasing power to accelerate deployment. It proposes specific institutional tasks (coordinator, pôle de compétences) and operational instruments (labels, guichets, indices measuring 'good complementarity' of AI and human work). While not legally binding, the framework aims to seed administrative capacity and build standard-setting practices that can later be translated into binding instruments where appropriate.

Monitoring and Evaluation

The report advocates a results-oriented monitoring and evaluation system combining quantitative metrics (training numbers, research outputs, compute capacity, startup growth, procurement outcomes) and qualitative evaluations (ethical audits, citizen evaluation exercises, sectoral experiment reviews). It proposes public dashboards and named accountability roles for the interministerial coordinator and recommends periodic reviews and mid-term evaluations for major infrastructure projects. The report also suggests piloting experimental metrics (e.g., indices of 'complementarity' between humans and AI in the workplace) and institutionalising public reporting of impact assessments for high-stakes systems.

Penalties, Liability, and Appeals

The Villani Report is primarily programmatic and does not itself create a new penalty regime. It emphasizes the need to clarify liability rules where AI systems cause harm and to ensure existing legal tools (civil liability, sectoral regulation, GDPR enforcement) are compatible with AI-induced risks. It recommends development of audit capacities and discrimination impact assessments to inform liability allocation and potential regulatory interventions. Where necessary, the report supports targeted legislative work to adjust labour law and liability frameworks, but it leaves specific sanctions and administrative penalties to subsequent legal processes and sectoral lawmaking. In data-protection contexts, the report points stakeholders toward existing enforcement under the GDPR and the Commission nationale de l'informatique et des libertés (CNIL).

Relationship to Other Instruments

The Villani Report situates itself relative to pre-existing national and European instruments: it builds on earlier French work (e.g., France IA) and ethics reports from CNIL and other advisory bodies, and it anticipates and aligns with emergent EU-level initiatives (ethics guidelines, later legislative proposals). It recommends that French policy both comply with GDPR and actively contribute to European standard-setting. The report's proposals (3IA, compute resources, sandboxes) were designed to complement research funding mechanisms, procurement law and sectoral regulatory regimes rather than to supersede them. It thus functions principally as a strategic, cross-cutting policy roadmap rather than as a source of direct regulatory mandates.

International Alignment

The report repeatedly stresses the need for European and international coordination, arguing that many AI challenges (data flows, compute, ethics, standardisation) transcend national borders and require multinational cooperation. Villani recommends France promote an "offensive" European data policy, collaborate on shared infrastructure and push for ambitious European timelines on sectoral authorisation (e.g., for autonomous vehicles). It proposes that France use diplomatic channels and research partnerships to influence global standards and to support European sovereignty in key areas (defence, critical industrial components, cloud and data centres). The report thus frames national measures as complementary to EU-level regulation and international norms.

Implementation Timeline

Sources and References