The Court of Auditors has just published an exhaustive analysis of the national strategy for artificial intelligence. While France has become one of the European leaders in terms of research, infrastructure and innovation in AI, the Court of Auditors estimates that the State itself remains one of the actors furthest behind in the adoption of these technologies. This gap, which is increasing as the private sector accelerates, directly threatens the performance of the public service and the ability of the administration to keep pace with technological transformations.
The institution recalls that France now ranks third in the world in training and research in AI and that it has established itself as the leading European ecosystem in generative AI. At the same time, more than a thousand startups active in this field, the extension of the Jean Zay supercomputer and the rise of academic clusters demonstrate undeniable scientific and entrepreneurial dynamism. However, despite these assets, the State has not modernized at the same pace. The Court notes that, between 2023 and 2025, no specific budget was dedicated to the transformation of public action through AI, even though this objective was at the heart of the second phase of the national strategy.
The report highlights that the first phase of the SNIA, between 2018 and 2022, made it possible to finance several pilot projects intended to test the use of AI in public missions such as fraud detection, automation of procedures or the exploitation of administrative data sets. But these initiatives remained confined to local experiments, often carried out by isolated teams, and were never deployed on a national scale. The Court insists on the absence of structuring effects, namely that the projects were carried out without an overall vision, without lasting funding and without articulation with the digital reforms carried out in the ministries.
Deprived of specific resources during the recent period, Dinum has sought to compensate by developing low-cost internal tools, such as the Albert API or the Compar:IA platform. These tools have made it possible to occasionally introduce automation or textual analysis capabilities, but their performance remains uneven. Certain experiments carried out in the France Services network have also highlighted problematic limits, particularly in terms of quality of responses, reliability of models and traceability of sources. The Court emphasizes that this strategy, based on internalization and prototypes, does not allow for credible scaling up or uniform adoption across administrations.
One of the most striking findings of the report concerns the very limited use of public procurement, yet at the heart of the innovation policies of many countries. The French State, notes the Court, has not mobilized its purchasing power to support a sovereign ecosystem of AI solutions nor to accelerate the industrialization of technologies resulting from experiments. This inertia places administrations in a fragile position: they remain dependent on insufficiently efficient internal solutions, while private actors are moving forward at high speed.
It also warns of the social and organizational consequences of this delay. She cites a study indicating that 38% of public jobs in France will be significantly impacted by generative AI in the coming years. This upheaval requires strategic management and anticipation of business transformations, otherwise the State risks finding itself out of step with its own agents and users. The risk is no longer that of a simple technological delay but that of a functional break in the capacity of the administration to carry out its missions.
To remedy these weaknesses, the Court makes a central recommendation to create a General Secretariat for Artificial Intelligence, reporting directly to the Prime Minister, responsible for coordinating all public actors, allocating the necessary resources and imposing a pace of execution compatible with the rapid evolution of technologies. This renewed governance must be accompanied by an increase in the power of dedicated teams in each ministry, a much more systematic use of existing AI solutions, a rigorous evaluation of their uses and massive support for the public officials concerned.
The Court finally calls for a change in logic in public action, by making the adoption of AI a lever for the quality and efficiency of public policies, and considers it essential to prioritize projects capable of significantly improving the fight against fraud, the automation of repetitive tasks or the fluidity of user journeys. It also invites the State to mobilize much more actively innovative public procurement, in order to support French solutions and guarantee a level of performance consistent with the needs of administrations.
In drawing up this report, the Court of Auditors recalls that France today has all the components of European leadership in AI. It now remains to align public action with this ambition. Without a rapid acceleration in the adoption of AI in administrations, the State risks becoming one of the last actors to benefit from the transformations it is calling for.
