Knowledge poisoning poses a critical threat to Retrieval-Augmented Generation (RAG) systems by injecting adversarial content into knowledge bases, tricking Large Language Models (LLMs) into producing attacker-controlled outputs grounded in manipulated context. Prior work highlights LLMs' susceptibility to misleading or malicious retrieved content. However, real-world fact-checking scenarios are more challenging, as credible evidence typically dominates the retrieval pool. To investigate this problem, we extend knowledge poisoning to the fact-checking setting, where retrieved context includes authentic supporting or refuting evidence. We propose \textbf{ADMIT} (\textbf{AD}versarial \textbf{M}ulti-\textbf{I}njection \textbf{T}echnique), a few-shot, semantically aligned poisoning attack that flips fact-checking decisions and induces deceptive justifications, all without access to the target LLMs, retrievers, or token-level control. Extensive experiments show that ADMIT transfers effectively across 4 retrievers, 11 LLMs, and 4 cross-domain benchmarks, achieving an average attack success rate (ASR) of 86\% at an extremely low poisoning rate of $0.93 \times 10^{-6}$, and remaining robust even in the presence of strong counter-evidence. Compared with prior state-of-the-art attacks, ADMIT improves ASR by 11.2\% across all settings, exposing significant vulnerabilities in real-world RAG-based fact-checking systems.

This paper discusses the potential for integrating Generative Artificial Intelligence (GenAI) into professional heritage practice with the aim of enhancing the accessibility of public-facing guidance documents. We developed HAZEL, a GenAI chatbot fine-tuned to assist with revising written guidance relating to heritage conservation and interpretation. Using quantitative assessments, we compare HAZEL's performance to that of ChatGPT (GPT-4) in a series of tasks related to the guidance writing process. The results of this comparison indicate a slightly better performance of HAZEL over ChatGPT, suggesting that the GenAI chatbot is more effective once the underlying large language model (LLM) has been fine-tuned. However, we also note significant limitations, particularly in areas requiring cultural sensitivity and more advanced technical expertise. These findings suggest that, while GenAI cannot replace human heritage professionals in technical authoring tasks, its potential to automate and expedite certain aspects of guidance writing could offer valuable benefits to heritage organisations, especially in resource-constrained contexts.

As Generative Artificial Intelligence is adopted across the financial services industry, a significant barrier to adoption and usage is measuring model performance. Historical machine learning metrics can oftentimes fail to generalize to GenAI workloads and are often supplemented using Subject Matter Expert (SME) Evaluation. Even in this combination, many projects fail to account for various unique risks present in choosing specific metrics. Additionally, many widespread benchmarks created by foundational research labs and educational institutions fail to generalize to industrial use. This paper explains these challenges and provides a Risk Assessment Framework to allow for better application of SME and machine learning Metrics

This survey provides the first systematic review of Arabic LLM benchmarks, analyzing 40+ evaluation benchmarks across NLP tasks, knowledge domains, cultural understanding, and specialized capabilities. We propose a taxonomy organizing benchmarks into four categories: Knowledge, NLP Tasks, Culture and Dialects, and Target-Specific evaluations. Our analysis reveals significant progress in benchmark diversity while identifying critical gaps: limited temporal evaluation, insufficient multi-turn dialogue assessment, and cultural misalignment in translated datasets. We examine three primary approaches: native collection, translation, and synthetic generation discussing their trade-offs regarding authenticity, scale, and cost. This work serves as a comprehensive reference for Arabic NLP researchers, providing insights into benchmark methodologies, reproducibility standards, and evaluation metrics while offering recommendations for future development.

Large language models (LLMs) are increasingly deployed across high-impact domains, from clinical decision support and legal analysis to hiring and education, making fairness and bias evaluation before deployment critical. However, existing evaluations lack grounding in real-world scenarios and do not account for differences in harm severity, e.g., a biased decision in surgery should not be weighed the same as a stylistic bias in text summarization. To address this gap, we introduce HALF (Harm-Aware LLM Fairness), a deployment-aligned framework that assesses model bias in realistic applications and weighs the outcomes by harm severity. HALF organizes nine application domains into three tiers (Severe, Moderate, Mild) using a five-stage pipeline. Our evaluation results across eight LLMs show that (1) LLMs are not consistently fair across domains, (2) model size or performance do not guarantee fairness, and (3) reasoning models perform better in medical decision support but worse in education. We conclude that HALF exposes a clear gap between previous benchmarking success and deployment readiness.

Large Reasoning Models (LRMs) excel at complex reasoning but are traditionally evaluated in static, "frozen world" settings: model responses are assumed to be instantaneous, and the context of a request is presumed to be immutable over the duration of the response. While generally true for short-term tasks, the "frozen world" assumption breaks down in modern reasoning tasks such as assistive programming, where models may take hours to think through problems and code may change dramatically from the time the model starts thinking to the model's final output. In this work, we challenge the frozen world assumption and evaluate LRM robustness under two realistic dynamic scenarios: interruptions, which test the quality of the model's partial outputs on a limited budget, and dynamic context, which tests model adaptation to in-flight changes. Across mathematics and programming benchmarks that require long-form reasoning, static evaluations consistently overestimate robustness: even state-of-the-art LRMs, which achieve high accuracy in static settings, can fail unpredictably when interrupted or exposed to changing context, with performance dropping by up to 60% when updates are introduced late in the reasoning process. Our analysis further reveals several novel failure modes, including reasoning leakage, where models fold the reasoning into their final answer when interrupted; panic, where under time pressure models abandon reasoning entirely and return incorrect answers; and self-doubt, where performance degrades while incorporating updated information. Project Page: http://dynamic-lm.github.io/

We present Entropic Mutual-Information Geometry Large-Language Model Alignment (ENIGMA), a novel approach to Large-Language Model (LLM) training that jointly improves reasoning, alignment and robustness by treating an organisation's policies/principles as directions to move on a model's information manifold. Our single-loop trainer combines Group-Relative Policy Optimisation (GRPO), an on-policy, critic-free RL method with Chain-of-Thought (CoT)-format only rewards; a Self-Supervised Alignment with Mutual Information (SAMI)-style symmetric InfoNCE auxiliary; and an entropic Sinkhorn optimal-transport regulariser on hidden-state distributions to bound geometry drift. We also introduce infoNCE metrics that specialise to a standard MI lower bound under matched negatives to measure how strongly a model's CoT encodes these policies. These metrics include a Sufficiency Index (SI) that enables the selection and creation of principles that maximise downstream performance prior to training. In our experiments using small (1B) LLMs, high-SI principles predict steadier training dynamics and improved benchmark performance over GRPO ablations. Our information-geometry analysis of trained models validates desirable structural change in the manifold. These results support our hypothesis that reasoning, alignment, and robustness are projections of a single information-geometric objective, and that models trained using ENIGMA demonstrate principled reasoning without the use of a reward model, offering a path to trusted capability

Detecting harmful AI actions is important as AI agents gain adoption. Chain-of-thought (CoT) monitoring is one method widely used to detect adversarial attacks and AI misalignment. However, attackers and misaligned models might evade CoT monitoring through ciphered reasoning: reasoning hidden in encrypted, translated, or compressed text. To assess this risk, we test whether models can perform ciphered reasoning. For each of 28 different ciphers, we fine-tune and prompt up to 10 models to reason in that cipher. Across the models we test, we find an asymmetry: model accuracy can drop significantly when reasoning in ciphered text, even though models demonstrate comprehension of ciphered text by being able to translate it accurately to English. Even frontier models struggle with lesser-known ciphers, although they can reason accurately in well-known ciphers like rot13. We show that ciphered reasoning capability correlates with cipher prevalence in pretraining data. We also identify scaling laws showing that ciphered reasoning capability improves slowly with additional fine-tuning data. Our work suggests that evading CoT monitoring using ciphered reasoning may be an ineffective tactic for current models and offers guidance on constraining the development of this capability in future frontier models. Modern large language models (LLMs) rely on chain-of-thought (CoT) (Wei et al., 2022) to achieve strong performance (Guo et al., 2025). CoT increases the proportion of model computation that occurs in natural language (Korbak et al., 2025), which allows automated systems to monitor model CoTs for misaligned behavior. CoT monitoring has been employed in tasks as diverse as reinforcement learning (RL) training of frontier models (Baker et al., 2025), AI control (Kutasov et al., 2025), frontier model evaluation (METR, 2025), agent monitoring (Meinke et al., 2024), and jailbreak safeguards (Sharma et al., 2025) to supervise model behavior. In many of these settings, access to legible reasoning traces is critical; without it, monitors are much less capable (Baker et al., 2025).

Large language models (LLMs) that fluently converse with humans are a reality - but do LLMs experience human-like processing difficulties? We systematically compare human and LLM sentence comprehension across seven challenging linguistic structures. We collect sentence comprehension data from humans and five families of state-of-the-art LLMs, varying in size and training procedure in a unified experimental framework. Our results show LLMs overall struggle on the target structures, but especially on garden path (GP) sentences. Indeed, while the strongest models achieve near perfect accuracy on non-GP structures (93.7% for GPT-5), they struggle on GP structures (46.8% for GPT-5). Additionally, when ranking structures based on average performance, rank correlation between humans and models increases with parameter count. For each target structure, we also collect data for their matched baseline without the difficult structure. Comparing performance on the target vs. baseline sentences, the performance gap observed in humans holds for LLMs, with two exceptions: for models that are too weak performance is uniformly low across both sentence types, and for models that are too strong the performance is uniformly high. Together, these reveal convergence and divergence in human and LLM sentence comprehension, offering new insights into the similarity of humans and LLMs.

We introduce PETLP (Privacy-by-design Extract, Transform, Load, and Present), a compliance framework that embeds legal safeguards directly into extended ETL pipelines. Central to PETLP is treating Data Protection Impact Assessments as living documents that evolve from preregistration through dissemination. Through systematic Red-dit analysis, we demonstrate how extraction rights fundamentally differ between qualifying research organisations (who can invoke DSM Article 3 to override platform restrictions) and commercial entities (bound by terms of service), whilst GDPR obligations apply universally. We demonstrate why true anonymisation remains unachievable for social media data and expose the legal gap between permitted dataset creation and uncertain model distribution. By structuring compliance decisions into practical workflows and simplifying institutional data management plans, PETLP enables researchers to navigate regulatory complexity with confidence, bridging the gap between legal requirements and research practice.