Aton Kamanda

I work on AI safety at LawZero, the nonprofit founded by Yoshua Bengio, where we are building Scientist AI: a non-agentic system that aims to understand and explain the world rather than act in it, so that it can serve as a trustworthy guardrail for more autonomous AI.

Before that, I spent my time turning research into working systems: LLM agents for legal translation at Alexa Translations, and real-time video understanding at AwakeAI, a Mila-incubated startup. I studied artificial intelligence at Mila and the Université de Montréal, and computer science at the University of Namur.

Publications3
  1. Safety from Honesty in a Disinterested AI Predictor
    Y. Bengio, O. Richardson, T. Gavenčiak, M. Cohen, …, A. Kamanda, A. Oberman, …, J. GhosnarXiv preprint, 2026
  2. Latent Personality Alignment: Improving Harmlessness Without Mentioning Harms
    L. Le, D. Williams-King, M. A. Merzouk, A. Kamanda, A. ObermanICLR 2026 Workshop · OpenReview
  3. CodeUltraFeedback: An LLM-as-a-Judge Dataset for Aligning LLMs to Coding Preferences
    M. Weyssow, A. Kamanda, X. Zhou, H. SahraouiACM, 2026 · ACM