Pierre Fernandez

I am a Research Scientist at FAIR Paris (Fundamental AI Research lab at Meta).

My research focuses on content protection and safety in machine learning, with a particular interest in watermarking generative models. My goal is to make content generated by AI traceable and identifiable. I also study how watermarks can reveal which data a model was trained on. I notably developed:

• TextSeal, a state-of-the-art watermark for LLMs, with stronger detection than SynthID-text, localization in mixed human/AI documents, and no inference overhead or quality loss (NeurIPS 2026, co-led with Tom Sander);
• the invisible watermarking of Meta’s AI-generated images, from Imagine to Muse Image, with a public detection tool at meta.ai/identification;
• Video Seal and Pixel Seal, state-of-the-art open-source image and video watermarking models, now part of Content Seal;
• AudioSeal, the audio watermarking algorithm used in public research demos from FAIR, such as Audiobox and Seamless, serving 100k users daily (joint work with Hady Elsahar);
• Stable Signature, a method to watermark generative models even if they are open-sourced.

I also co-organized the GenAI Watermarking workshop at ICLR 2025 and a tutorial on watermarking at CVPR 2026, and I am regularly consulted on the technical aspects of AI transparency regulations, such as the EU AI Act.

I did my PhD at FAIR and Inria Rennes, advised by Matthijs Douze, Hervé Jégou and Teddy Furon. Prior to my PhD I studied at École polytechnique, majoring in computer science and mathematics for visual computing. I also hold a Master’s degree from École des Ponts and from Paris-Saclay University where I studied Mathematics for vision and learning.