AI security · Cybersecurity · Trustworthy AI

Securing artificial intelligence for healthcare, defense and industry.

I study how AI systems are attacked and how to make them hold. Senior Scientist and Junior Professor at INSERM and IMT Atlantique (LaTIM, Brest), with earlier research at Inria, NIST and the University of Siena.

HDR · PhD · Executive MBA · IEEE Senior Member · 3 Springer books · 30+ peer-reviewed papers · 1,300+ citations

Why it matters

Most attacks on AI do not break anything. A backdoored model still scores 99% on every test you run, until it meets the trigger it was trained to obey.

My work is to find these weaknesses and close them before they reach a hospital, a mission or a product. More than a decade of adversarial research across France, the USA and Italy.

Get in touch

Let’s talk about your AI risk, your event or your research

Advisory, keynotes, research collaboration or a senior role: the Contact page has a short form that comes straight to me.

Get in touch

Work with me

Three ways we can work together

Organizations and boards: advisory on AI risk

Find what AI risk looks like in your own stack
Secure health-data platforms and AI agents
Prepare for NIS2, the Cyber Resilience Act and the AI Act
Now: Scientific Advisor to BRENUS PHARMA and board member at ViDiTrust

Events and media: keynotes and talks

For clinicians, CISOs, engineers and executives
Keynote at the Sorbonne/CEA Machine Learning Security Workshop
Hospitals on FHIR: Berlin 2025, Pisa 2026
In English, French, Italian or Arabic

Research partners: collaboration and supervision

Joint projects on secure, robust and private AI
Director of two PhD theses on federated learning security in hospitals
Programs with Thales, SOPHiA GENETICS, CHU Brest and national labs
Editor, reviewer and program committee member

Leadership roles

I am open to senior industry leadership positions in AI security. Recruiters can reach me through the Contact page.

Focus areas

What I work on

AI security

How machine learning systems fail under attack, and how to measure it.

Backdoor and data-poisoning attacks

Hidden behaviors planted during training that standard tests do not reveal.

Federated learning security

Keeping models that hospitals train together robust to malicious participants.

Model IP protection

DNN watermarking that proves who owns a trained model, even when only its outputs are visible.

LLM, RAG and agentic AI

Privacy for clinical records and secure integration of AI agents.

AI governance

Turning technical risk into decisions that boards and regulators can act on.

Research

Selected results

PARIS: robust federated learning

IEEE FLTA 2026, 2nd Best Paper Award. A defense that keeps federated learning robust when some participants send malicious or corrupted updates, designed with hospital federations in mind.

GHOST: backdoors that persist in federated learning

IEEE Transactions on Information Forensics and Security, 2026. Shows how one malicious participant can keep a hidden behavior alive when hospital data differ from site to site. Knowing this is the first step to defenses that hold.

RoSe and Mixer: proving who owns a model

IEEE TDSC 2026 and ICASSP 2023 (Top 3% paper). Black-box watermarking that lets a model owner verify ownership from the model outputs alone, with security that can be measured.

A backdoor attack without label poisoning

IEEE ICIP 2019, 700+ citations. One of the first clean-label backdoor attacks on CNNs: the training labels stay correct, so the poisoning is hard to spot.