Trustworthy AI in defense: our contribution to the PEC white paper

The Pôle d'excellence cyber publishes the third edition of its white paper. Our contribution focuses on a major issue: the auditing of AI models. LLMs are indeed “cultural objects”, propagating the biases of their designers and their training data, aka the internet.

Today, the rise of AI agents reinforces this threat by making invisible the impact that a model aligned to certain values can have. And yet, models are now both upstream AND downstream of crises. A biased model can trigger one, while others can amplify existing crises...

Our approach to this risk? Using AI to audit AI, while keeping humans in the loop: agents specialized in analyzing potential drifts around large bases of vulnerabilities and biases, metrics and comparison data, for an assisted panoptic view and informed “realignment” recommendations.

To find out more, drop us a line or read our article in the white paper available here.