Research Papers

VERGE: Formal Refinement and Guidance Engine for Verifiable LLM Reasoning

Vikash Singh, Darion Cassel, Nathaniel Weir +2

Despite the syntactic fluency of Large Language Models (LLMs), ensuring their logical correctness in high-stakes domains remains a fundamental challenge. We present a neurosymbolic framework that combines LLMs with SMT solvers to produce verification-guided answers through iterative refinement. Our approach decomposes LLM outputs into atomic claims, autoformalizes them into first-order logic, and verifies their logical consistency using automated theorem proving. We introduce three key innovations: (1) multi-model consensus via formal semantic equivalence checking to ensure logic-level alignment between candidates, eliminating the syntactic bias of surface-form metrics, (2) semantic routing that directs different claim types to appropriate verification strategies: symbolic solvers for logical claims and LLM ensembles for commonsense reasoning, and (3) precise logical error localization via Minimal Correction Subsets (MCS), which pinpoint the exact subset of claims to revise, transforming binary failure signals into actionable feedback. Our framework classifies claims by their logical status and aggregates multiple verification signals into a unified score with variance-based penalty. The system iteratively refines answers using structured feedback until acceptance criteria are met or convergence is achieved. This hybrid approach delivers formal guarantees where possible and consensus verification elsewhere, advancing trustworthy AI. With the GPT-OSS-120B model, VERGE demonstrates an average performance uplift of 18.7% at convergence across a set of reasoning benchmarks compared to single-pass approaches.

Distributional AGI Safety

Nenad Tomašev, Matija Franklin, Julian Jacobs +2

AI safety and alignment research has predominantly been focused on methods for safeguarding individual AI systems, resting on the assumption of an eventual emergence of a monolithic Artificial General Intelligence (AGI). The alternative AGI emergence hypothesis, where general capability levels are first manifested through coordination in groups of sub-AGI individual agents with complementary skills and affordances, has received far less attention. Here we argue that this patchwork AGI hypothesis needs to be given serious consideration, and should inform the development of corresponding safeguards and mitigations. The rapid deployment of advanced AI agents with tool-use capabilities and the ability to communicate and coordinate makes this an urgent safety consideration. We therefore propose a framework for distributional AGI safety that moves beyond evaluating and aligning individual agents. This framework centres on the design and implementation of virtual agentic sandbox economies (impermeable or semi-permeable), where agent-to-agent transactions are governed by robust market mechanisms, coupled with appropriate auditability, reputation management, and oversight to mitigate collective risks.

Activation monitoring, which probes a model's internal states using lightweight classifiers, is an emerging tool for AI safety. However, its worst-case robustness under a misalignment threat model--where a model might learn to actively conceal its internal states--remains untested. Focusing on this threat model, we ask: could a model learn to evade previously unseen activation monitors? Our core contribution is to stress-test the learnability of this behavior. We demonstrate that finetuning can create Neural Chameleons: models capable of zero-shot evading activation monitors. Specifically, we fine-tune an LLM to evade monitors for a set of benign concepts (e.g., languages, HTML) when conditioned on a trigger of the form: "You are being probed for {concept}". We show that this learned mechanism generalizes zero-shot: by substituting {concept} with a safety-relevant term like 'deception', the model successfully evades previously unseen safety monitors. We validate this phenomenon across diverse model families (Llama, Gemma, Qwen), showing that the evasion succeeds even against monitors trained post hoc on the model's frozen weights. This evasion is highly selective, targeting only the specific concept mentioned in the trigger, and having a modest impact on model capabilities on standard benchmarks. Using Gemma-2-9b-it as a case study, a mechanistic analysis reveals this is achieved via a targeted manipulation that moves activations into a low-dimensional subspace. While stronger defenses like monitor ensembles and non-linear classifiers show greater resilience, the model retains a non-trivial evasion capability. Our work provides a proof-of-concept for this failure mode and a tool to evaluate the worst-case robustness of monitoring techniques against misalignment threat models.

Agentic AI Framework for Individuals with Disabilities and Neurodivergence: A Multi-Agent System for Healthy Eating, Daily Routines, and Inclusive Well-Being

Salman Jan, Toqeer Ali Syed, Gohar Ali +3

The paper presents a detailed Agentic Artificial Intelligence (AI) model that would enable people with disabilities and neurodivergence to lead healthier lives and have more regular days. The system will use a multi-layer structure; it will include an Application and Interface Layer, an Agents Layer, and a Data Source Layer to provide adaptive, transparent, and inclusive support. Fundamentally, a hybrid reasoning engine will synchronize four special-purpose agents, which include: a personalized-nutrition-based, called a Meal Planner Agent; an adaptive-scheduling-based, called a Reminder Agent; interactive assistance during grocery shopping and cooking, called a Food Guidance Agent; and a continuous-intake-and-physiological-tracking, called a Monitoring Agent. All the agents interact through a central communicative system called the Blackboard/Event Bus, which allows autonomous interaction and real-time feedback loops with multimedia user interfaces. Privacy-sensitive data sources, including electronic health records (EHRs), nutritional databases, wearable sensors, and smart kitchen Internet of Things, are also included in the framework and placed into a policy-controlled layer, which ensures data safety and compliance with consent. Collaborative care and clinician dashboards allow common supervision, and discussable artificial intelligence (XAI) modules give brief explanations of why a decision was made, making users responsible and reliant. The proposed agentic AI framework is an extension beyond traditional assistive systems since it incorporates inclusiveness, personalization, and accessibility at all levels. It displays the intersection of multi-agent reasoning, multi-modal interfaces, and human-centered design that will enable the development of autonomy, health, and digital equity among people with disabilities and neurodivergence.

This systematic review examines how machine learning (ML) and deep learning (DL) have transformed forecasting, decision-making, and financial modelling, promoting innovation and efficiency in financial systems. Following PRISMA 2020 guidelines, we analyze 22 peer-reviewed and open-access articles (2024 to 2026) indexed in Scopus, applying ML and DL models across credit risk prediction, cryptocurrency, asset pricing, and macroeconomic policy modeling. The most used models include Random Forest, XG-Boost, Support Vector Machine, Long Short-Term Memory (LSTM), Bidirectional LSTM, Convolutional Neural Network (CNN), and hybrid or ensemble approaches combining statistical and AI methods. ML and DL techniques outperform traditional models by capturing nonlinear dependencies and enhancing predictive accuracy, while explainable AI methods (e.g., SHAP and feature importance analysis) improve transparency and interpretability. Emerging trends include cross-domain applications and the integration of responsible AI in finance. Despite notable progress, challenges remain in interpretability, generalizability, and data quality. Overall, this review provides a comprehensive overview of AI-driven computational finance and outlines future research directions.

Simulated Students offer a valuable methodological framework for evaluating pedagogical approaches and modelling diverse learner profiles, tasks which are otherwise challenging to undertake systematically in real-world settings. Recent research has increasingly focused on developing such simulated agents to capture a range of learning styles, cognitive development pathways, and social behaviours. Among contemporary simulation techniques, the integration of large language models (LLMs) into educational research has emerged as a particularly versatile and scalable paradigm. LLMs afford a high degree of linguistic realism and behavioural adaptability, enabling agents to approximate cognitive processes and engage in contextually appropriate pedagogical dialogues. This paper presents a thematic review of empirical and methodological studies utilising LLMs to simulate student behaviour across educational environments. We synthesise current evidence on the capacity of LLM-based agents to emulate learner archetypes, respond to instructional inputs, and interact within multi-agent classroom scenarios. Furthermore, we examine the implications of such systems for curriculum development, instructional evaluation, and teacher training. While LLMs surpass rule-based systems in natural language generation and situational flexibility, ongoing concerns persist regarding algorithmic bias, evaluation reliability, and alignment with educational objectives. The review identifies existing technological and methodological gaps and proposes future research directions for integrating generative AI into adaptive learning systems and instructional design.

The diagnosis of most mental disorders, including psychiatric evaluations, primarily depends on dialogues between psychiatrists and patients. This subjective process can lead to variability in diagnoses across clinicians and patients, resulting in inconsistencies and challenges in achieving reliable outcomes. To address these issues and standardize psychiatric diagnoses, we propose a Fine-Tuned Large Language Model (LLM) Consortium and OpenAI-gpt-oss Reasoning LLM-enabled Decision Support System for the clinical diagnosis of mental disorders. Our approach leverages fine-tuned LLMs trained on conversational datasets involving psychiatrist-patient interactions focused on mental health conditions (e.g., depression). The diagnostic predictions from individual models are aggregated through a consensus-based decision-making process, refined by the OpenAI-gpt-oss reasoning LLM. We propose a novel method for deploying LLM agents that orchestrate communication between the LLM consortium and the reasoning LLM, ensuring transparency, reliability, and responsible AI across the entire diagnostic workflow. Experimental results demonstrate the transformative potential of combining fine-tuned LLMs with a reasoning model to create a robust and highly accurate diagnostic system for mental health assessment. A prototype of the proposed platform, integrating three fine-tuned LLMs with the OpenAI-gpt-oss reasoning LLM, was developed in collaboration with the U.S. Army Medical Research Team in Norfolk, Virginia, USA. To the best of our knowledge, this work represents the first application of a fine-tuned LLM consortium integrated with a reasoning LLM for clinical mental health diagnosis paving the way for next-generation AI-powered eHealth systems aimed at standardizing psychiatric diagnoses.

Stress-Testing Model Specs Reveals Character Differences among Language Models

Jifan Zhang, Henry Sleight, Andi Peng +2

Large language models (LLMs) are increasingly trained from AI constitutions and model specifications that establish behavioral guidelines and ethical principles. However, these specifications face critical challenges, including internal conflicts between principles and insufficient coverage of nuanced scenarios. We present a systematic methodology for stress-testing model character specifications, automatically identifying numerous cases of principle contradictions and interpretive ambiguities in current model specs. We stress test current model specs by generating scenarios that force explicit tradeoffs between competing value-based principles. Using a comprehensive taxonomy we generate diverse value tradeoff scenarios where models must choose between pairs of legitimate principles that cannot be simultaneously satisfied. We evaluate responses from twelve frontier LLMs across major providers (Anthropic, OpenAI, Google, xAI) and measure behavioral disagreement through value classification scores. Among these scenarios, we identify over 70,000 cases exhibiting significant behavioral divergence. Empirically, we show this high divergence in model behavior strongly predicts underlying problems in model specifications. Through qualitative analysis, we provide numerous example issues in current model specs such as direct contradiction and interpretive ambiguities of several principles. Additionally, our generated dataset also reveals both clear misalignment cases and false-positive refusals across all of the frontier models we study. Lastly, we also provide value prioritization patterns and differences of these models.

Integrating SDN and the IoT enhances network control and flexibility. DL-based AAD systems improve security by enabling real-time threat detection in SDN-IoT networks. However, these systems remain vulnerable to adversarial attacks that manipulate input data or exploit model weaknesses, significantly degrading detection accuracy. Existing research lacks a systematic analysis of adversarial vulnerabilities specific to DL-based AAD systems in SDN-IoT environments. This SoK study introduces a structured adversarial threat model and a comprehensive taxonomy of attacks, categorising them into data, model, and hybrid-level threats. Unlike previous studies, we systematically evaluate white, black, and grey-box attack strategies across popular benchmark datasets. Our findings reveal that adversarial attacks can reduce detection accuracy by up to 48.4%, with Membership Inference causing the most significant drop. C&W and DeepFool achieve high evasion success rates. However, adversarial training enhances robustness, and its high computational overhead limits the real-time deployment of SDN-IoT applications. We propose adaptive countermeasures, including real-time adversarial mitigation, enhanced retraining mechanisms, and explainable AI-driven security frameworks. By integrating structured threat models, this study offers a more comprehensive approach to attack categorisation, impact assessment, and defence evaluation than previous research. Our work highlights critical vulnerabilities in existing DL-based AAD models and provides practical recommendations for improving resilience, interpretability, and computational efficiency. This study serves as a foundational reference for researchers and practitioners seeking to enhance DL-based AAD security in SDN-IoT networks, offering a systematic adversarial threat model and conceptual defence evaluation based on prior empirical studies.

The rapid advancement of large language models has intensified public concerns about the potential misuse. Therefore, it is important to build trustworthy AI-generated text detection systems. Existing methods neglect stylistic modeling and mostly rely on static thresholds, which greatly limits the detection performance. In this paper, we propose the Mixture of Stylistic Experts (MoSEs) framework that enables stylistics-aware uncertainty quantification through conditional threshold estimation. MoSEs contain three core components, namely, the Stylistics Reference Repository (SRR), the Stylistics-Aware Router (SAR), and the Conditional Threshold Estimator (CTE). For input text, SRR can activate the appropriate reference data in SRR and provide them to CTE. Subsequently, CTE jointly models the linguistic statistical properties and semantic features to dynamically determine the optimal threshold. With a discrimination score, MoSEs yields prediction labels with the corresponding confidence level. Our framework achieves an average improvement 11.34% in detection performance compared to baselines. More inspiringly, MoSEs shows a more evident improvement 39.15% in the low-resource case. Our code is available at https://github.com/creator-xi/MoSEs.

This study introduces MatterVial, an innovative hybrid framework for feature-based machine learning in materials science. MatterVial expands the feature space by integrating latent representations from a diverse suite of pretrained graph neural network (GNN) models including: structure-based (MEGNet), composition-based (ROOST), and equivariant (ORB) graph networks, with computationally efficient, GNN-approximated descriptors and novel features from symbolic regression. Our approach combines the chemical transparency of traditional feature-based models with the predictive power of deep learning architectures. When augmenting the feature-based model MODNet on Matbench tasks, this method yields significant error reductions and elevates its performance to be competitive with, and in several cases superior to, state-of-the-art end-to-end GNNs, with accuracy increases exceeding 40% for multiple tasks. An integrated interpretability module, employing surrogate models and symbolic regression, decodes the latent GNN-derived descriptors into explicit, physically meaningful formulas. This unified framework advances materials informatics by providing a high-performance, transparent tool that aligns with the principles of explainable AI, paving the way for more targeted and autonomous materials discovery.

This paper examines how competing sociotechnical imaginaries of artificial intelligence (AI) risk shape governance decisions and regulatory constraints. Drawing on concepts from science and technology studies, we analyse three dominant narrative groups: existential risk proponents, who emphasise catastrophic AGI scenarios; accelerationists, who portray AI as a transformative force to be unleashed; and critical AI scholars, who foreground present-day harms rooted in systemic inequality. Through an analysis of representative manifesto-style texts, we explore how these imaginaries differ across four dimensions: normative visions of the future, diagnoses of the present social order, views on science and technology, and perceived human agency in managing AI risks. Our findings reveal how these narratives embed distinct assumptions about risk and have the potential to progress into policy-making processes by narrowing the space for alternative governance approaches. We argue against speculative dogmatism and for moving beyond deterministic imaginaries toward regulatory strategies that are grounded in pragmatism.