Can AI draft new laws? A study tests simulated constitutional assemblies

Benjamin Ricaud and Artem Galushko studied language-model agents debating a hypothetical constitution. Their preprint reports that agents combine familiar legal concepts with AI-specific proposals, but struggle to reason through the deeper consequences of novel situations. This is an experimental study, not a test of an enacted law or proof that AI can replace legislators. Human legal reasoning remains central to the authors’ conclusions.