REFLECTION NOTE · STRATÉGIE & IA DE DÉFENSE · 2026

Governing defence AI.

Technical analysis and governance proposal for defense AI. Consensus-M architecture, inspired by theTrias Politica.

Sacha Virgile Elouardi · Designer · Open-source AI research · Multi-agent architecture

The black-box trap

The scenario raised by the French Ministry of the Armed Forces Red Team highlights a critical risk in modern warfare: dependence on a monolithic AI whose decision-making path is opaque. In a high-intensity combat environment, corrupted data—whether caused by software, electronic warfare or systemic bias—could lead an AI to make a catastrophic decision without any human operator being able to understand its root cause. Without traceability, command is left with an unacceptable explanatory gap, both operationally and legally.

The Asymmetric Multi-Agent Concept (CMAA)

The alternative is to reject centralized AI in favor of a separation-of-powers structure directly inspired by military doctrine and public law. This model relies on three autonomous, specialized agents that are isolated from one another.

POWER TO ACT

Operational AI

A purely tactical analysis of the terrain. It maps the environment, identifies the forces present (enemies, civilians and allies) and determines the optimal technical approach for achieving the military objective.

POWER OF OVERSIGHT

Legal & Ethical AI

Completely separate, it receives the Operational AI plan and checks it against rules of engagement, international humanitarian law (IHL) and mission history. It calculates an objective benefit-risk ratio that takes past failures and collateral damage into account.

REAL-TIME VALIDATION

Tactical Fire AI

As the final link in the kinetic chain, it validates the order immediately before the effect, continuously comparing the analyses of the first two AIs with the dynamic situation in the fraction of a second preceding the action.

The Dissensus Score

The technological core of the architecture is divergence detection. Instead of seeking forced consensus, the system measures a mathematical index of decision variance in real time: the Dissensus Score, denoted Sd.

Sd > τ

If the Operational AI and Legal AI analyses diverge beyond the tolerable critical threshold because of an anomaly or corrupted data, the architecture triggers aFail-Safe: autonomy is suspended and the decision is transferred to human command, according to the principleHuman-in-the-loop.

Designer context

This proposal is part of an active, pragmatic research process. I am currently developing an artificial-intelligence architecture based on an open-source model. Drawing on this hands-on experience with open code and language models, I believe that code transparency, combined with rigorous multi-agent engineering, is a key way to reduce the black-box opacity that legitimately concerns the defense industry.

Sacha Virgile Elouardi · Designer · Open-source AI research · Multi-agent architecture.

▶

VIDEO COMING SOON

An accessible presentation of this analysis will be published on YouTube soon.

Does this analysis raise questions for you? Would you like to challenge it, add to it or explore a collaboration?

DISCUSS THIS TOPIC