France Publishes the First Volume of a Practical Guide to Help Public Buyers Purchase AI
France’s public buyers have a new reference for purchasing artificial intelligence. The first volume of a practical guide to AI procurement, drawn up under the aegis of the Conseil national de la commande publique (CNCP), was published on November 6, the government’s digital portal numerique.gouv.fr announced. It gives administrations benchmarks and concrete tools to understand the AI market and prepare purchases before a tender is even launched.
The guide starts from the opportunities AI offers public services, such as automating tasks, analysing data and making services more efficient, but it stresses that buying such systems raises distinct issues: data protection, information-system security, regulatory compliance, control over how the technology evolves and environmental impact. Because AI tools differ widely in how they are built and deployed, buyers are told they need a clear grasp of data security, reversibility and control of the solutions they choose.
In practice, the first volume helps buyers map the stakes around protecting and securing their data, identify the main ways of implementing AI systems and the points of vigilance attached to each, and adapt their sourcing, the market-research phase in which they talk to potential suppliers, to their own needs and constraints. Coverage by Fides Algorithmica highlighted the weight given to reversibility, meaning an administration’s ability to switch provider or technology without being locked into a system it cannot leave.
What it means in practice
The document was written collaboratively by several central directorates, including the interministerial digital directorate DINUM, the legal affairs directorate DAJ and the state purchasing directorate DAE, with input from expert bodies such as the CNIL and ANSSI, local-authority representatives, digital-industry federations and purchasing organisations. A second volume dedicated to the contract clauses that should be written into deals with AI suppliers is due at the end of November 2026.
The guidance formalises questions that public bodies already face, from Twake.ai’s recommendation as a sovereign workplace for more than 2,500 public bodies to the Banque de France’s plan to run Mistral AI on a sovereign cloud. What to watch is how quickly ministries and local authorities apply the sourcing checklist, what the contract-clause volume says about data use, exit terms and liability, and whether suppliers adjust their offers to the reversibility requirements.