Within a legal framework, fairness in datasets and models is typically assessed by dividing observations into predefined groups and then computing fairness measures (e.g., Disparate Impact or Equality of Odds with respect to gender). However, when sensitive attributes such as skin color are continuous, dividing into default groups may overlook or obscure the discrimination experienced by certain minority subpopulations. To address this limitation, we propose a fairness-based grouping approach for continuous (possibly multidimensional) sensitive attributes. By grouping data according to observed levels of discrimination, our method identifies the partition that maximizes a novel criterion based on inter-group variance in discrimination, thereby isolating the most critical subgroups. We validate the proposed approach using multiple synthetic datasets and demonstrate its robustness under changing population distributions - revealing how discrimination is manifested within the space of sensitive attributes. Furthermore, we examine a specialized setting of monotonic fairness for the case of skin color. Our empirical results on both CelebA and FFHQ, leveraging the skin tone as predicted by an industrial proprietary algorithm, show that the proposed segmentation uncovers more nuanced patterns of discrimination than previously reported, and that these findings remain stable across datasets for a given model. Finally, we leverage our grouping model for debiasing purpose, aiming at predicting fair scores with group-by-group post-processing. The results demonstrate that our approach improves fairness while having minimal impact on accuracy, thus confirming our partition method and opening the door for industrial deployment.

The optimization of industrial processes remains a critical challenge, particularly when no mathematical formulation of objective functions or constraints is available. This study addresses this issue by proposing a surrogate-based, data-driven methodology for optimizing complex real-world manufacturing systems using only historical process data. Machine learning models are employed to approximate system behavior and construct surrogate models, which are integrated into a tailored metaheuristic approach: Data-Driven Differential Evolution with Multi-Level Penalty Functions and Surrogate Models, an adapted version of Differential Evolution suited to the characteristics of the studied process. The methodology is applied to an extrusion process in the tire manufacturing industry, with the goal of optimizing initialization parameters to reduce waste and production time. Results show that the surrogate-based optimization approach outperforms historical best configurations, achieving a 65\% reduction in initialization and setup time, while also significantly minimizing material waste. These findings highlight the potential of combining data-driven modeling and metaheuristic optimization for industrial processes where explicit formulations are unavailable.

Vision-language models (VLMs) are increasingly applied to identify unsafe or inappropriate images due to their internal ethical standards and powerful reasoning abilities. However, it is still unclear whether they can recognize various unsafe concepts when presented in different modalities, such as text and images. To address this, we first compile the UnsafeConcepts dataset, featuring 75 unsafe concepts, i.e., ``Swastika,'' ``Sexual Harassment,'' and ``Assaults,'' along with associated 1.5K images. We then conduct a systematic evaluation of VLMs' perception (concept recognition) and alignment (ethical reasoning) capabilities. We assess eight popular VLMs and find that, although most VLMs accurately perceive unsafe concepts, they sometimes mistakenly classify these concepts as safe. We also identify a consistent modality gap among open-source VLMs in distinguishing between visual and textual unsafe concepts. To bridge this gap, we introduce a simplified reinforcement learning (RL)-based approach using proximal policy optimization (PPO) to strengthen the ability to identify unsafe concepts from images. Our approach uses reward scores based directly on VLM responses, bypassing the need for collecting human-annotated preference data to train a new reward model. Experimental results show that our approach effectively enhances VLM alignment on images while preserving general capabilities. It outperforms baselines such as supervised fine-tuning (SFT) and direct preference optimization (DPO). We hope our dataset, evaluation findings, and proposed alignment solution contribute to the community's efforts in advancing safe VLMs.

Large Language Models (LLMs) can memorize and reveal personal information, raising concerns regarding compliance with the EU's GDPR, particularly the Right to Be Forgotten (RTBF). Existing machine unlearning methods assume the data to forget is already known but do not address how to identify which individual-fact associations are stored in the model. Privacy auditing techniques typically operate at the population level or target a small set of identifiers, limiting applicability to individual-level data inquiries. We introduce WikiMem, a dataset of over 5,000 natural language canaries covering 243 human-related properties from Wikidata, and a model-agnostic metric to quantify human-fact associations in LLMs. Our approach ranks ground-truth values against counterfactuals using calibrated negative log-likelihood across paraphrased prompts. We evaluate 200 individuals across 15 LLMs (410M-70B parameters), showing that memorization correlates with subject web presence and model scale. We provide a foundation for identifying memorized personal data in LLMs at the individual level, enabling the dynamic construction of forget sets for machine unlearning and RTBF requests.

Diffusion-based large language models (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs, offering faster inference and greater interactivity via parallel decoding and bidirectional modeling. However, despite strong performance in code generation and text infilling, we identify a fundamental safety concern: existing alignment mechanisms fail to safeguard dLLMs against context-aware, masked-input adversarial prompts, exposing novel vulnerabilities. To this end, we present DIJA, the first systematic study and jailbreak attack framework that exploits unique safety weaknesses of dLLMs. Specifically, our proposed DIJA constructs adversarial interleaved mask-text prompts that exploit the text generation mechanisms of dLLMs, i.e., bidirectional modeling and parallel decoding. Bidirectional modeling drives the model to produce contextually consistent outputs for masked spans, even when harmful, while parallel decoding limits model dynamic filtering and rejection sampling of unsafe content. This causes standard alignment mechanisms to fail, enabling harmful completions in alignment-tuned dLLMs, even when harmful behaviors or unsafe instructions are directly exposed in the prompt. Through comprehensive experiments, we demonstrate that DIJA significantly outperforms existing jailbreak methods, exposing a previously overlooked threat surface in dLLM architectures. Notably, our method achieves up to 100% keyword-based ASR on Dream-Instruct, surpassing the strongest prior baseline, ReNeLLM, by up to 78.5% in evaluator-based ASR on JailbreakBench and by 37.7 points in StrongREJECT score, while requiring no rewriting or hiding of harmful content in the jailbreak prompt. Our findings underscore the urgent need for rethinking safety alignment in this emerging class of language models. Code is available at https://github.com/ZichenWen1/DIJA.

This paper presents our approach to EXIST 2025 Task 1, addressing text-based sexism detection in English and Spanish tweets through hierarchical Low-Rank Adaptation (LoRA) of Llama 3.1 8B. Our method introduces conditional adapter routing that explicitly models label dependencies across three hierarchically structured subtasks: binary sexism identification, source intention detection, and multilabel sexism categorization. Unlike conventional LoRA applications that target only attention layers, we apply adaptation to all linear transformations, enhancing the model's capacity to capture task-specific patterns. In contrast to complex data processing and ensemble approaches, we show that straightforward parameter-efficient fine-tuning achieves strong performance. We train separate LoRA adapters (rank=16, QLoRA 4-bit) for each subtask using unified multilingual training that leverages Llama 3.1's native bilingual capabilities. The method requires minimal preprocessing and uses standard supervised learning. Our multilingual training strategy eliminates the need for separate language-specific models, achieving 1.7-2.4\% F1 improvements through cross-lingual transfer. With only 1.67\% trainable parameters compared to full fine-tuning, our approach reduces training time by 75\% and model storage by 98\%, while achieving competitive performance across all subtasks (ICM-Hard: 0.6774 for binary classification, 0.4991 for intention detection, 0.6519 for multilabel categorization).

AI models need to be unlearned to fulfill the requirements of legal acts such as the AI Act or GDPR, and also because of the need to remove toxic content, debiasing, the impact of malicious instances, or changes in the data distribution structure in which a model works. Unfortunately, removing knowledge may cause undesirable side effects, such as a deterioration in model performance. In this paper, we investigate the problem of adversarial unlearning, where a malicious party intentionally sends unlearn requests to deteriorate the model's performance maximally. We show that this phenomenon and the adversary's capabilities depend on many factors, primarily on the backbone model itself and strategy/limitations in selecting data to be unlearned. The main result of this work is a new method of protecting model performance from these side effects, both in the case of unlearned behavior resulting from spontaneous processes and adversary actions.

Argumentation frameworks (AFs) provide formal approaches for legal reasoning, but identifying sources of ambiguity and explaining argument acceptance remains challenging for non-experts. We present AF-XRAY, an open-source toolkit for exploring, analyzing, and visualizing abstract AFs in legal reasoning. AF-XRAY introduces: (i) layered visualizations based on game-theoretic argument length revealing well-founded derivation structures; (ii) classification of attack edges by semantic roles (primary, secondary, blunders); (iii) overlay visualizations of alternative 2-valued solutions on ambiguous 3-valued grounded semantics; and (iv) identification of critical attack sets whose suspension resolves undecided arguments. Through systematic generation of critical attack sets, AF-XRAY transforms ambiguous scenarios into grounded solutions, enabling users to pinpoint specific causes of ambiguity and explore alternative resolutions. We use real-world legal cases (e.g., Wild Animals as modeled by Bench-Capon) to show that our tool supports teleological legal reasoning by revealing how different assumptions lead to different justified conclusions.

Conventional anti-money laundering (AML) systems predominantly focus on identifying anomalous entities or transactions, flagging them for manual investigation based on statistical deviation or suspicious behavior. This paradigm, however, misconstrues the true nature of money laundering, which is rarely anomalous but often deliberate, repeated, and concealed within consistent behavioral routines. In this paper, we challenge the entity-centric approach and propose a network-theoretic perspective that emphasizes detecting predefined laundering patterns across directed transaction networks. We introduce the notion of behavioral consistency as the core trait of laundering activity, and argue that such patterns are better captured through subgraph structures expressing semantic and functional roles - not solely geometry. Crucially, we explore the concept of pattern fragility: the sensitivity of laundering patterns to small attribute changes and, conversely, their semantic robustness even under drastic topological transformations. We claim that laundering detection should not hinge on statistical outliers, but on preservation of behavioral essence, and propose a reconceptualization of pattern similarity grounded in this insight. This philosophical and practical shift has implications for how AML systems model, scan, and interpret networks in the fight against financial crime.

Unstructured text from legal, medical, and administrative sources offers a rich but underutilized resource for research in public health and the social sciences. However, large-scale analysis is hampered by two key challenges: the presence of sensitive, personally identifiable information, and significant heterogeneity in structure and language. We present a modular toolchain that prepares such text data for embedding-based analysis, relying entirely on open-weight models that run on local hardware, requiring only a workstation-level GPU and supporting privacy-sensitive research. The toolchain employs large language model (LLM) prompting to standardize, summarize, and, when needed, translate texts to English for greater comparability. Anonymization is achieved via LLM-based redaction, supplemented with named entity recognition and rule-based methods to minimize the risk of disclosure. We demonstrate the toolchain on a corpus of 10,842 Swedish court decisions under the Care of Abusers Act (LVM), comprising over 56,000 pages. Each document is processed into an anonymized, standardized summary and transformed into a document-level embedding. Validation, including manual review, automated scanning, and predictive evaluation shows the toolchain effectively removes identifying information while retaining semantic content. As an illustrative application, we train a predictive model using embedding vectors derived from a small set of manually labeled summaries, demonstrating the toolchain's capacity for semi-automated content analysis at scale. By enabling structured, privacy-conscious analysis of sensitive documents, our toolchain opens new possibilities for large-scale research in domains where textual data was previously inaccessible due to privacy and heterogeneity constraints.