Two long papers accepted at ACL 2026!

Two of our papers have been accepted as long papers in the main track at ACL 2026 (Annual Meeting of the Association for Computational Linguistics), to be held in San Diego, California.

---

LLMs (Almost) Never Abstain Under Medical Uncertainty
A. Cocchieri, L. Ragazzi, G. Tagliavini, G. Moro

We introduce MedQAbstain, a new evaluation framework that assesses how language models handle medical decision-making when appropriate answers are unknown. State-of-the-art LLMs systematically overcommit, rarely abstaining even when the question itself is hidden — revealing a significant gap between model behaviour and clinical safety requirements.

---

Sycophants in the Courtroom: Are LLMs Fragile to Juridical Authority and Evolving Legal Standards?
L. Molfetta, A. Cocchieri, L. Ragazzi, I. Bartolini, M. Patella, G. Moro

We examine how well language models reason over legal text, proposing that law differs fundamentally from medicine in its reliance on authoritative sources and temporal validity. Our findings show that models treat law as unstructured text rather than binding precedent, demonstrating weakness in assessing citation relevance and susceptibility to formatting changes.