Challenge

Generative LLMs are already used to draft legal opinions, petitions and memoranda. One of the most documented risks is source hallucination: the model cites case law or legal provisions that do not exist, or does so too vaguely to be verified. In recent years, courts in Brazil and abroad have already sanctioned lawyers for filing briefs with fabricated, AI-generated citations.

The challenge involves building the automatic verifier this workflow is missing: a system that reads a legal opinion, identifies every case-law and statutory citation, and classifies each one as real, fabricated or incomplete for verification.