Newest papers

Commit or Abstain? System One Decision Models, LLMs, and Missing Legal Authority

M. Mikail Demir · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29961 · cited by 0 works (OpenAlex) · open-access PDF

French

Compte tenu de l'énorme nombre de revendications à vérifier, l'IA juridique vérifiable nécessite des modèles évolutifs et rentables qui répondent aux besoins pratiques de la loi. Les modèles de décision System One sont construits sur de grands modèles linguistiques mais, au lieu de générer du texte, renvoient une probabilité calibré pour chaque option. Cela convient à la vérification de l'autorité juridique, où la question est fermée: une détention donnée, ou aucune d'entre elles. Nous testons si de tels modèles s'abstiennent lorsque l'autorité correcte manque. Nous étendons CaseHOLD avec une condition d'abstention dans laquelle la détention correcte est remplacée par 1,670 des 5,314 éléments, et comparons quatre

Machine translation (Meta M2M100); the English original follows.

Abstract

Given the enormous number of claims to be checked, verifiable legal AI requires scalable, cost-effective models that answer the practical needs of law. System One decision models are built on large language models but, instead of generating text, return a calibrated probability for each option. This suits the verification of legal authority, where the question is closed: a given holding, or none of them. We test whether such models abstain when the correct authority is missing. We extend CaseHOLD with an abstention condition in which the correct holding is replaced in 1,670 of 5,314 items, and compare four decision models with six generative LLMs. When no option is correct, generative LLMs still pick one in 38-86% of items; the most cautious decision model does so in 30%. On standard CaseHOLD, a decision model matches the best generative LLM (88.1% vs. 88.2%) at 1/55 of the cost. Accepting that LLM's answer only when a decision model agrees halves its wrong citations, and distilling a decision model's probabilities into Qwen3-8B, without labels, yields a well-calibrated local model. Legal AI should be judged by its wrong citations, not by accuracy alone.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Commit, Computer science, Generative grammar, Decision model, Verifiable secret sharing, Decision analysis

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange