Defence AI Agency Amiad Details How French Forces Will Use Mistral Models, From Contracts to Pendragon Robots
France’s defence AI agency, Amiad, has given one of its most detailed public accounts of how the armed forces plan to use models from Mistral AI, in an interview with Numerama published on October 31. Alain Droniou, technical deputy director of the agency, which was set up in spring 2024, described projects ranging from contract drafting to planning for Pendragon, the first French military robotics unit.
The framework agreement the ministry signed with Mistral in January 2026 gives the agency access to models reserved for enterprise customers, not just the open-weight ones anyone can download, as well as Mistral’s computing infrastructure and the start-up’s expertise. Droniou said that sensitive use cases, which are numerous, still run on the ministry’s own infrastructure.
Early applications include helping the Direction générale de l’armement draft contracts with major industrial suppliers by analysing past procurement records and automating steps from drafting to notification. Operational uses for the Navy and the Army centre on mission and strategic planning, such as logistics, resupply and force deployment for naval vessels. Droniou said Mistral models will also be used to build agent-based tools for planning and conducting Pendragon’s missions, alongside other technologies meant to guarantee compliance with the law of war and the rules of engagement. Support for targeting and the simulation of adversary reactions are among the prospective uses the agency intends to study.
What it means in practice
Amiad has a team of just over ten people assessing AI risk along five lines: vulnerabilities such as hallucinations or poisoned datasets, information-system security, human factors such as cognitive overload, operational safety, and legal and ethical risk. Droniou said no project has been stopped so far, but some have been changed, and critical systems may face penetration-style testing to see whether the AI can be made to fail. The interview fits a wider pattern of the state leaning on Mistral, after the government awarded Mistral €6 million in ministry deals while warning about monopoly risk, and of security work such as Orange Cyberdefense’s partnership with Mistral on AI penetration testing.
Droniou said agentic AI has gained a great deal of capability over the past six to twelve months, and the ministry wants an agent orchestration platform on which tools can be added and swapped, sometimes after six months, instead of software projects that run for two or three years.
What to watch next is when the first contract-drafting tools reach procurement staff, how Pendragon’s agent tools are tested before use in the field, and whether the ministry publishes its risk framework.