People
Who is doing this.
Who is here, and — just as plainly — which seats are still empty. An initiative that hides its size is spending credibility it has not earned.
Founding researcher
Pierre Lague
Pierre Lague · Founding researcherPierre Lague is a French AI researcher and PhD student at Inria Rennes (I4S team) and the Université d'Angers, in collaboration with the Direction Générale de l'Armement. His doctoral research focuses on physics-informed machine learning for structural health monitoring and vibration characterisation, with a particular emphasis on open-source, reproducible implementations of standards-based methodologies — work where the reliability and interpretability of AI systems are non-negotiable.
He holds a Master's degree in Machine Learning (Université de Lille — ranked 3rd, with honours) and in Complex Systems Engineering, and a Bachelor's in Computer Science (Université Bretagne Sud — class major). Throughout his studies he built research experience through multiple internships at the CNRS, leading projects in semantic segmentation for aerial drone imagery and 3D point-cloud classification.
He then joined the private sector as a Machine Learning Engineer, building the entire technological stack for cloud optimisation with deep-learning time-series forecasting and full autonomy over research and product development. His broader technical interests span reinforcement learning, multi-agent systems, explainable AI, and physics-informed neural networks — the last of which is where the initiative's insistence on measurable, checkable claims comes from.
My research has always lived in the same place: making powerful models you can actually trust — physics-informed, reproducible, and explainable enough that an expert can verify them rather than merely believe them.
Agent governance is currently the opposite of that. There are strong positions everywhere and almost no measurements. People propose registries without knowing whether an unregistered agent can be detected, and revocation without knowing how long revocation takes. Those are empirical questions, and nobody is answering them.
That is why I started this initiative — and why it is a community rather than a company. Nobody settles these questions alone, and nobody should trust the party selling the answer.
Trajectory
A track record in reliable, interpretable AI.
PhD Researcher — Physics-Informed ML
Inria Rennes (I4S) · Université d'Angers · with the DGA
Doctoral research on physics-informed machine learning for structural health monitoring and vibration characterisation — with a focus on open-source, reproducible, standards-based implementations, where the reliability and interpretability of AI systems are non-negotiable.
Machine Learning Engineer
Sudo Group
Built the entire technological stack for cloud optimisation using deep-learning time-series forecasting, with full autonomy over both research and product development.
Research Internships — Computer Vision
CNRS
Led projects in semantic segmentation for aerial drone imagery and 3D point-cloud classification across multiple internships.
MSc Machine Learning · MSc Complex Systems · BSc Computer Science
Université de Lille · Université Bretagne Sud · Université Technologique de Compiègne
Master's in Machine Learning (Université de Lille — ranked 3rd, with honours) and Complex Systems Engineering; Bachelor's in Computer Science (Université Bretagne Sud — class major).
Open seats
The rest of this page is vacancies.
Stated as vacancies rather than dressed up as a roster. Every one of these is currently unfilled, and filling them well matters more than filling them fast.
Affiliated researchers
One per agenda question, eventually. They keep their own post and affiliation, carry a question, and publish under their own name.
Working-group chairs
A chair convenes a track, keeps the question honest, and is responsible for the group publishing something — including a documented dead end.
Reviewers
Named adversarial readers. Nothing is published without at least one review from someone with standing to say it does not hold.
Advisory circle
Law, regulation, and public administration, to keep the technical work attached to instruments that actually exist. Advisory, never editorial.
Editorial rules
How anything gets published.
Authors decide what a note says
The initiative can decline to publish under its name; it cannot change a conclusion. A note the initiative disagrees with is published with the disagreement attached, signed.
Nothing ships without adversarial review
At least one named reviewer with standing to refuse. Reviews that stop a draft are recorded, because the record of what we did not publish matters as much as the record of what we did.
Interests are declared before the work, not after
Employment, funding, advisory roles, and equity, stated at the top of each note. A member with an interest in an outcome can still work on it — they just cannot do so quietly.
Corrections are versioned, never silent
When we are shown to be wrong, the note is amended in place with a dated correction and the original left readable. An initiative that edits its past is not a research body.
Put your name on this page.
Affiliated researchers, chairs, reviewers, students, and institutional interlocutors. No exclusivity, named authorship, and the freedom to publish a conclusion we do not like.