Etienne Salimbeni | Searcharxiv

arXiv2026

LEXam: Benchmarking Legal Reasoning on 340 Law Exams

Long-form legal reasoning remains a key challenge for large language models (LLMs) in spite of recent advances in test-time scaling. To address this, we introduce LEXam, a novel benchmark derived from 340 law exams spanning 116 law school courses across a range of subjects and degree levels. The dataset comprises 7,537 law exam questions in English and German. It includes both long-form, open-ended questions and multiple-choice questions with varying numbers of options. Besides reference answers, the open questions are also accompanied by explicit guidance outlining the expected legal reasoning approach such as issue spotting, rule recall, or rule application. Our evaluation on both open-ended and multiple-choice questions present significant challenges for current LLMs; in particular, they notably struggle with open questions that require structured, multi-step legal reasoning. Moreover, our results underscore the effectiveness of the dataset in differentiating between models with varying capabilities. Deploying an ensemble LLM-as-a-Judge paradigm with rigorous human expert validation, we demonstrate how model-generated reasoning steps can be evaluated consistently and accurately, closely aligning with human expert assessments. Our evaluation setup provides a scalable method to assess legal reasoning quality beyond simple accuracy metrics. Project page: https://lexam-benchmark.github.io/.

cs.CL↗

arXiv2023

Your Email Address Holds the Key: Understanding the Connection Between Email and Password Security with Deep Learning

In this work, we investigate the effectiveness of deep-learning-based password guessing models for targeted attacks on human-chosen passwords. In recent years, service providers have increased the level of security of users'passwords. This is done by requiring more complex password generation patterns and by using computationally expensive hash functions. For the attackers this means a reduced number of available guessing attempts, which introduces the necessity to target their guess by exploiting a victim's publicly available information. In this work, we introduce a context-aware password guessing model that better capture attackers'behavior. We demonstrate that knowing a victim's email address is already critical in compromising the associated password and provide an in-depth analysis of the relationship between them. We also show the potential of such models to identify clusters of users based on their password generation behaviour, which can spot fake profiles and populations more vulnerable to context-aware guesses. The code is publicly available at https://github.com/spring-epfl/DCM_sp

cs.CR↗