Jailbreaking large language models (LLMs) has emerged as a pressing concern with the increasing prevalence and accessibility of conversational LLMs. Adversarial users often exploit these models through carefully engineered prompts to elicit restricted or sensitive outputs, a strategy widely referred to as jailbreaking. While numerous defense mechanisms have been proposed, attackers continuously develop novel prompting techniques, and no existing model can be considered fully resistant. In this study, we investigate the jailbreak phenomenon by examining the internal representations of LLMs, with a focus on how hidden layers respond to jailbreak versus benign prompts. Specifically, we analyze the open-source LLM GPT-J and the state-space model Mamba2, presenting preliminary findings that highlight distinct layer-wise behaviors. Our results suggest promising directions for further research on leveraging internal model dynamics for robust jailbreak detection and defense.

As large language models (LLMs) are deployed globally, creating pluralistic systems that can accommodate the diverse preferences and values of users worldwide becomes essential. We introduce EVALUESTEER, a benchmark to measure LLMs' and reward models' (RMs) steerability towards users' value and stylistic preference profiles grounded in psychology and human-LLM interaction literature. To address the gap in existing datasets that do not support controlled evaluations of RM steering, we synthetically generated 165,888 preference pairs -- systematically varying pairs along 4 value dimensions (traditional, secular-rational, survival, and self-expression) and 4 style dimensions (verbosity, readability, confidence, and warmth). We use EVALUESTEER to evaluate whether, given a user profile and a pair of candidate value-laden and style-laden responses, LLMs and RMs are able to select the output that aligns with the user's preferences. We evaluate six open-source and proprietary LLMs and RMs under eleven systematic prompting conditions and six preference comparison scenarios. Notably, our results show that, when given the user's full profile of values and stylistic preferences, the best models achieve <75% accuracy at choosing the correct response, in contrast to >99% accuracy when only relevant style and value preferences are provided. EVALUESTEER thus highlights the limitations of current RMs at identifying and adapting to relevant user profile information, and provides a challenging testbed for developing RMs that can be steered towards diverse human values and preferences.

Abstract--As Large Language Models (LLMs) gain wider adoption, ensuring their reliable handling of Personally Identifiable Information (PII) across diverse regulatory contexts has become essential. This work introduces a scalable multilingual data curation framework designed for high-quality PII annotation across 13 underrepresented locales (Table I), covering approximately 336 locale-specific PII types. Our phased, human-in-the-loop annotation methodology combines linguistic expertise with rigorous quality assurance, leading to substantial improvements in recall and false positive rates from pilot, training, and production phases. Beyond reporting empirical gains, we highlight common annotator challenges in multilingual PII labeling and demonstrate how iterative, analytics-driven pipelines can enhance both annotation quality and downstream model reliability. I. Introduction A. PII Data Protection The surge in user-generated content has led to vast textual corpora containing hidden instances of Personally Identifiable Information (PII) in application forms, support tickets, reviews and social media posts [1]. PII--such as NAME, SSN, and PHONE NUMBER--poses significant privacy risks if not handled correctly. Its compromise can result in identity theft, financial fraud, and unauthorized access to sensitive data [2].

Identifying disease interconnections through manual analysis of large-scale clinical data is labor-intensive, subjective, and prone to expert disagreement. While machine learning (ML) shows promise, three critical challenges remain: (1) selecting optimal methods from the vast ML landscape, (2) determining whether real-world clinical data (e.g., electronic health records, EHRs) or structured disease descriptions yield more reliable insights, (3) the lack of "ground truth," as some disease interconnections remain unexplored in medicine. Large language models (LLMs) demonstrate broad utility, yet they often lack specialized medical knowledge. Our framework integrates the following: (i) a statistical co-occurrence analysis and a masked language modeling (MLM) approach using real clinical data; (ii) domain-specific BERT variants (Med-BERT and BioClinicalBERT); (iii) a general-purpose BERT and document retrieval; and (iv) four LLMs (Mistral, DeepSeek, Qwen, and Y andexGPT). Our graph-based comparison of the obtained interconnection matrices shows that the LLM-based approach produces interconnections with the lowest diversity of ICD code connections to different diseases compared to other methods, including text-based and domain-based approaches. This suggests an important implication: LLMs have limited potential for discovering new interconnections. In the absence of ground truth databases for medical interconnections between ICD codes, our results constitute a valuable medical disease ontology that can serve as a founda-tional resource for future clinical research and artificial intelligence applications in healthcare. Electronic health records (EHRs) provide a valuable resource for studying disease progression and relationships between diagnoses. Machine learning (ML) can help discover hidden patterns in medical data, but many existing models are hard to interpret. In particular, it is not always clear whether large language models (LLMs) make predictions based on meaningful medical knowledge or simply rely on textual similarities between diagnosis descriptions (Cui et al., 2025). This is especially critical in healthcare, where model decisions must align with established medical knowledge and pathophysiological mechanisms. We also analyze and compare the obtained results and summarize it into medical disease ontology.

Abstract--Efficient scheduling of large language model (LLM) inference tasks is essential for achieving low latency and high throughput, particularly with the growing use of reasoning-capable LLMs. Traditional strategies like First Come, First-Serve (FCFS) often suffer from Head-of-Line (HOL) blocking, where long-running tasks delay shorter ones queued behind them. In this paper, we introduce PARS, a prompt-aware LLM task scheduler that improves serving efficiency by approximating shortest-job-first (SJF) scheduling through pairwise ranking with margin ranking loss. PARS focuses on impactful scheduling decisions and seamlessly integrates into the state-of-the-art LLM serving system vLLM. It effectively predicts response-length-based task ordering, reducing latency with minimal overhead. Extensive experiments across multiple LLMs and real-world inference datasets show that PARS significantly improves performance, including for reasoning workloads. Furthermore, our cross-model evaluations demonstrate that the design generalizes well, enabling effective scheduling even when predictors are trained on different LLMs. Large language models (LLMs) have emerged as core engines for artificial intelligence applications, demonstrating remarkable capabilities in a wide range of tasks, including question answering, code generation, and text classification.

Large Language Models (LLMs) suffer from a range of vulnerabilities that allow malicious users to solicit undesirable responses through manipulation of the input text. These so-called jailbreak prompts are designed to trick the LLM into circumventing the safety guardrails put in place to keep responses acceptable to the developer's policies. In this study, we analyse the ability of different machine learning models to distinguish jailbreak prompts from genuine uses, including looking at our ability to identify jailbreaks that use previously unseen strategies. Our results indicate that using current datasets the best performance is achieved by fine tuning a Bidirectional Encoder Representations from Transformers (BERT) model end-to-end for identifying jailbreaks. We visualise the keywords that distinguish jailbreak from genuine prompts and conclude that explicit reflexivity in prompt structure could be a signal of jailbreak intention.

The study of emergent behaviors in large language model (LLM)-driven multi-agent systems is a critical research challenge, yet progress is limited by a lack of principled methodologies for controlled experimentation. To address this, we introduce Shachi, a formal methodology and modular framework that decomposes an agent's policy into core cognitive components: Configuration for intrinsic traits, Memory for contextual persistence, and Tools for expanded capabilities, all orchestrated by an LLM reasoning engine. This principled architecture moves beyond brittle, ad-hoc agent designs and enables the systematic analysis of how specific architectural choices influence collective behavior. We validate our methodology on a comprehensive 10-task benchmark and demonstrate its power through novel scientific inquiries. Critically, we establish the external validity of our approach by modeling a real-world U.S. tariff shock, showing that agent behaviors align with observed market reactions only when their cognitive architecture is appropriately configured with memory and tools. Our work provides a rigorous, open-source foundation for building and evaluating LLM agents, aimed at fostering more cumulative and scientifically grounded research.

Large Language Models (LLMs) have demonstrated impressive capabilities across various tasks, but fine-tuning them for domain-specific applications often requires substantial domain-specific data that may be distributed across multiple organizations. Federated Learning (FL) offers a privacy-preserving solution, but faces challenges with computational constraints when applied to LLMs. Low-Rank Adaptation (LoRA) has emerged as a parameter-efficient fine-tuning approach, though a single LoRA module often struggles with heterogeneous data across diverse domains. This paper addresses two critical challenges in federated LoRA fine-tuning: 1. determining the optimal number and allocation of LoRA experts across heterogeneous clients, and 2. enabling clients to selectively utilize these experts based on their specific data characteristics. We propose FedLEASE (Federated adaptive LoRA Expert Allocation and SElection), a novel framework that adaptively clusters clients based on representation similarity to allocate and train domain-specific LoRA experts. It also introduces an adaptive top-$M$ Mixture-of-Experts mechanism that allows each client to select the optimal number of utilized experts. Our extensive experiments on diverse benchmark datasets demonstrate that FedLEASE significantly outperforms existing federated fine-tuning approaches in heterogeneous client settings while maintaining communication efficiency.

Large language models (LLMs) are increasingly deployed as evaluators of text quality, yet the validity of their judgments remains underexplored. This study investigates systematic bias in self- and cross-model evaluations across three prominent LLMs: ChatGPT, Gemini, and Claude. We designed a controlled experiment in which blog posts authored by each model were evaluated by all three models under four labeling conditions: no attribution, true attribution, and two false-attribution scenarios. Evaluations employed both holistic preference voting and granular quality ratings across three dimensions Coherence, Informativeness, and Conciseness with all scores normalized to percentages for direct comparison. Our findings reveal pronounced asymmetries in model judgments: the "Claude" label consistently elevated scores regardless of actual authorship, while the "Gemini" label systematically depressed them. False attribution frequently reversed preference rankings, producing shifts of up to 50 percentage points in voting outcomes and up to 12 percentage points in quality ratings. Notably, Gemini exhibited severe self-deprecation under true labels, while Claude demonstrated intensified self-preference. These results demonstrate that perceived model identity can substantially distort both high-level judgments and fine-grained quality assessments, independent of content quality. Our findings challenge the reliability of LLM-as-judge paradigms and underscore the critical need for blind evaluation protocols and diverse multi-model validation frameworks to ensure fairness and validity in automated text evaluation and LLM benchmarking.

The effective training of Large Language Models (LLMs) for function calling faces a critical challenge: balancing exploration of complex reasoning paths with stable policy optimization. Standard methods like Supervised Fine-Tuning (SFT) fail to instill robust reasoning, and traditional Reinforcement Learning (RL) struggles with inefficient exploration. We propose \textbf{EGPO}, a new RL framework built upon Group Relative Policy Optimization (GRPO), designed to address this challenge directly. The core of EGPO is an entropy-enhanced advantage function that integrates the entropy of the model's Chain-of-Thought (CoT) into the policy gradient computation. This encourages the generation of diverse reasoning strategies. To maintain optimization direction, the entropy bonus is carefully constrained by a clipping mechanism. Complemented by a strict, binary reward signal, EGPO effectively guides the model towards discovering structured and accurate tool invocation patterns. On the challenging Berkeley Function Calling Leaderboard (BFCL), a 4B-parameter model trained with EGPO sets a new state-of-the-art among models of comparable size, surpassing a range of strong competitors, including GPT-4o and Gemini-2.5.