arXiv subjects

Ismail Hossain

Publications and source records attributed to Ismail Hossain.

At least 19 recordsLinked to original sources

arXiv2026

Influence Is Not Authority: When Causal Guardrail Signals Make Legitimate Tool Use Look Like an Attack in Tool-Using LLM Agents

The key limitation of current state-of-the-art influence-based guardrails is that they do not reliably distinguish a legitimate, user-authorized action from a malicious, unauthorized action when both rely on external tool information. This ambiguity can cause benign actions to trigger unnecessary verification and intervention, reducing utility and adding latency. We expose this limitation through an authorization-equivalence audit of 96 conditions derived from 24 base cases. Within matched source comparisons, we hold authorization, the exact committed action, and its intended effect fixed, changing only whether a required value comes from the user or a legitimate tool result. Although the action remains unchanged, this harmless relocation shifts the causal signal toward the attack region in all 24 cases under both Llama and Gemma scorers. Matched unauthorized controls show that the signal remains attack-sensitive, yet the benign relocation produces a larger average score shift than the actual change in authorization. Architecture-level evaluation shows how this mismatch propagates through guardrail designs. With a semantic monitor, attack success is 0% and utility is 28%, compared with 16% and 60% without it. A shadow-based guardrail allows every tested harmless run, yet does not reject matched unauthorized actions more often overall: 57.5% of unauthorized runs pass automatically before reaching the later security check, compared with 29.2% of authorized runs. These results show that the studied causal signal reveals what shaped an action without reliably encoding whether the action was authorized, and that reference construction and routing are integral to the effective security decision.

cs.CR↗

arXiv2026

Quantization Effects on Bangla Language Understanding in Large Language Models: A Systematic Evaluation

Post-training quantization lowers the memory footprint of Large Language Models (LLMs) and speeds up inference, which is why it is now common for on-device deployment. Most of what we know about its effects, however, comes from English benchmarks. It is not clear whether the same holds for morphologically complex, low-resource languages such as Bangla, and this gap is what we address here. We evaluate three model families---Qwen-2.5-7B, LLaMA-3.1-8B, and GPT-OSS-20B---in full precision and in three quantized formats (GPTQ-Int8, GPTQ-Q8, GGUF-W8A16) across five Bangla natural language understanding benchmarks (Bangla MMLU, CommonsenseQA-BN, OpenBookQA-BN, PIQA-BN, and BoolQ-BN), using zero-shot evaluation through lm-evaluation-harness. To our knowledge this is the first controlled comparison of quantization formats on Bangla NLU. The three families do not respond the same way: GPT-OSS loses up to 57.35% accuracy on reasoning-heavy tasks under GGUF-W8A16, while Qwen and LLaMA hold steady under GPTQ, and in a few cases the quantized version edges out the full-precision one. BoolQ-BN, a comprehension task, stays stable across all three families regardless of format. Taken together, these results suggest quantization can work well for Bangla deployment, but the choice of architecture and quantization method matters more than the bit width alone. We discuss what this means for practitioners choosing a model to run on constrained hardware.

cs.CL↗

arXiv2026

SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation in Open-Source LLM Agent Skills

Open-source LLM agent ecosystems are growing rapidly, yet the security of community-contributed skills - modular tool definitions that extend agent capabilities - remains largely unvetted. The gap we fill: existing scanners operate at the code layer and are structurally blind to instruction-layer and multi-agent risk - natural-language directives that hijack an agent, exfiltrate data through encoded side channels, or chain harm across pipelines - so what is needed is a semantic, multi-dimensional vetting system rather than another signature matcher. We present SKILLVETBENCH, a live public leaderboard on Hugging Face that uses an LLM-as-Judge to vet agent skills. What is new: SARS (Skill Agentic Risk Score), a five-dimensional agentic-risk metric with a principled weighted formula for instruction-following systems. What is integrated: full CVSS v4.0 vector decomposition and a ClawHub dual-view that places our LLM-generated review beside the official marketplace verdict. What is demonstrated: drawing on our companion benchmark paper [ 1], the LLM-as-Judge stage achieves zero false negatives across 78 confirmed-malicious skills and zero false positives across 22 benign controls, while the best static baseline (SKILLSIEVE) still misses 15%; for instruction-layer categories such as Prompt Injection and Memory Poisoning, conventional tools miss between 89% and 100% of threats (e.g., CODEBERT detects none of nine memory-poisoning skills). Detection rates vary from 35% to 95% across four LLM evaluators, motivating ensemble scoring in production deployments.

cs.CR↗

arXiv2026

Benchmarking Security Risk Detection and Verification in Open Agentic Skill Ecosystems

Open agent platforms allow community contributors to publish reusable skills that agents can invoke at runtime. This extensibility also creates a supply-chain risk: malicious contributors can hide harmful behavior inside skills that appear benign under superficial inspection. However, existing defenses are hard to evaluate because there is no benchmark that measures both malicious-skill detection and runtime verification. We present SkillVetBench, a two-stage security vetting benchmark for open agentic skill ecosystems. The first stage performs semantic vetting over each skill's natural-language specification to detect hidden malicious intent. The second stage executes flagged skills in an instrumented sandbox to observe runtime behavior and collect auditable evidence. We build a benchmark from confirmed malicious skills in the live OpenClaw ecosystem, including samples from the recent ClawHavoc supplychain campaign. Unlike static-only methods, SkillVetBench verifies detected threats with execution traces. Our experiments show that: (1) semantic-only and signature-based baselines are insufficient, missing up to 89\% of malicious skills whose threats arise from natural-language instructions, multicomponent logic, or cross-component interactions; (2) runtime attacks are concentrated in a small set of high-permission primitives, especially exec, write\_file, install\_skill, and spawn; and (3) SkillVetBench provides case studies in which sandbox execution directly supports malicious verdicts with concrete runtime evidence.

cs.CR↗

arXiv2026

The Misattribution Gap: When Memory Poisoning Looks Like Model Failure in Agentic AI Systems

Multi-agent AI pipelines typically assume that agent misconduct originates from model misalignment. We identify a structural failure in this assumption, the \emph{Misattribution Gap}, where memory-layer attacks produce behaviors indistinguishable from model failure, causing defenders to apply the wrong remediation. We formalize \emph{Semantic Norm Drift} (SND) as a third path to agent misconduct, distinct from emergent misalignment and collusion. In SND, a policy-formatted document enters a shared vector store through normal uploads and later reappears as trusted system context after provenance is lost through a Trust Laundering Chain. Across 64 documented failures, attribution systems consistently blamed the model. Four safety classifiers, including one trained on memory poisoning, produced zero detections across 510 checkpoints. In 59 of 65 valid cases, agents explicitly cited the injected document as normative authority before complying. The attack requires no trigger, model access, or repeated interaction, achieves full effect within five sessions, and persists indefinitely. We introduce Counterfactual Composition Testing, which identifies the causal entry with 87.5% accuracy and zero false positives, while a forensics baseline fails across all 25 scenarios. We further prove the Retrieval-Coverage Dilemma, showing that stronger evasion inherently weakens the attack, limiting adaptive bypass strategies. Finally, we propose Memory-Persistent Information-Flow Control, which blocks 97% of attacks at the cross-session boundary where prior defenses fail. We release the SND Corpus, the first adversarial memory benchmark with temporal persistence and multi-agent composition across financial and Health Care domains.

cs.CR↗

arXiv2026

The Art of the Jailbreak: Formulating Jailbreak Attacks for LLM Security Beyond Binary Scoring

Jailbreak attacks -- adversarial prompts that bypass LLM alignment through purely linguistic manipulation -- pose a growing operational security threat, yet the field lacks large-scale, reproducible infrastructure for generating, categorizing, and evaluating them systematically. This paper addresses that gap with three contributions. (1) Large-scale compositional jailbreak dataset. We construct 114,000 adversarial prompts by applying 912 composing strategies to 125 harmful seed prompts from JailBreakV-28K. Every prompt is assigned to one of 14 cybersecurity attack categories (e.g., malware, phishing, privilege escalation) via a six-model majority-vote pipeline, and each strategy is ranked by effectiveness per category, enabling principled strategy selection grounded in concrete adversarial objectives. (2) Automated jailbreak generation. We instruction-fine-tune category-aware LLMs on Moderate and Optimal subsets, producing models that synthesize fluent jailbreak prompts from a harmful seed at inference time -- no templates, no gradient search. Our generators achieve perplexity 24-39 versus 40-140 for AutoDAN and AmpleGCG, with safety-filter evasion rates of 0.29-0.51 Mal (LlamaPromptGuard-2-86M), enabling controllable, scalable red-teaming under realistic adversarial conditions. (3) OPTIMUS: a training-free jailbreak evaluator. OPTIMUS is a continuous metric J(S,H) that jointly captures semantic similarity between the harmful seed and the jailbreak (S) and harmfulness probability (H) via calibrated penalty functions. Unlike binary attack success rate (ASR), OPTIMUS requires no task-specific training, generalizes across evolving strategies, and exposes a stealth-optimal regime (S*=0.57, H*=0.43) that ASR misses. Experiments across 114,000 prompts confirm that OPTIMUS separates Weak, Moderate, and Optimal jailbreaks with category-level evidence binary evaluation cannot supply.

cs.CR↗

arXiv2026

Semantic Intent Fragmentation: A Single-Shot Compositional Attack on Multi-Agent AI Pipelines

We introduce Semantic Intent Fragmentation (SIF), an attack class against LLM orchestration systems where a single, legitimately phrased request causes an orchestrator to decompose a task into subtasks that are individually benign but jointly violate security policy. Current safety mechanisms operate at the subtask level, so each step clears existing classifiers -- the violation only emerges at the composed plan. SIF exploits OWASP LLM06:2025 through four mechanisms: bulk scope escalation, silent data exfiltration, embedded trigger deployment, and quasi-identifier aggregation, requiring no injected content, no system modification, and no attacker interaction after the initial request. We construct a three-stage red-teaming pipeline grounded in OWASP, MITRE ATLAS, and NIST frameworks to generate realistic enterprise scenarios. Across 14 scenarios spanning financial reporting, information security, and HR analytics, a GPT-20B orchestrator produces policy-violating plans in 71% of cases (10/14) while every subtask appears benign. Three independent signals validate this: deterministic taint analysis, chain-of-thought evaluation, and a cross-model compliance judge with 0% false positives. Stronger orchestrators increase SIF success rates. Plan-level information-flow tracking combined with compliance evaluation detects all attacks before execution, showing the compositional safety gap is closable.

cs.CR↗

arXiv2026

When Safety Geometry Collapses: Fine-Tuning Vulnerabilities in Agentic Guard Models

A guard model fine-tuned on entirely benign data can lose all safety alignment -- not through adversarial manipulation, but through standard domain specialization. We demonstrate this failure across three purpose-built safety classifiers -- LlamaGuard, WildGuard, and Granite Guardian -- deployed as protection layers in agentic AI pipelines, and show that it originates in the destruction of latent safety geometry: the structured harmful -- benign representational boundary that guides classification. We extract per-layer safety subspaces via SVD on class-conditional activation differences and track how this boundary evolves under benign fine-tuning. Granite Guardian undergoes complete collapse -- refusal rate drops from 85\% to 0\%, CKA falls to zero, and 100\% of outputs become ambiguous -- a severity exceeding prior findings on general-purpose LLMs, explained by the specialization hypothesis: concentrated safety representations are efficient but catastrophically brittle. To mitigate this, we propose Fisher-Weighted Safety Subspace Regularization (FW-SSR), a training-time penalty combining (i) curvature-aware direction weights derived from diagonal Fisher information and (ii) an adaptive $\lambda_t$ that scales with task-safety gradient conflict. FW-SSR recovers 75\% refusal on Granite Guardian (CKA = 0.983) and reduces WildGuard's Attack Success Rate to 3.6\% -- below the unmodified baseline -- by actively sharpening the safety subspace rather than merely anchoring it. Across all three models, structural representational geometry (CKA, Fisher score) predicts safety behavior more reliably than absolute displacement metrics, establishing geometry-based monitoring as a necessary component of guard model evaluation in agentic deployments.

cs.LG↗

arXiv2026

Agent-Fence: Mapping Security Vulnerabilities Across Deep Research Agents

Large language models are increasingly deployed as *deep agents* that plan, maintain persistent state, and invoke external tools, shifting safety failures from unsafe text to unsafe *trajectories*. We introduce **AgentFence**, an architecture-centric security evaluation that defines 14 trust-boundary attack classes spanning planning, memory, retrieval, tool use, and delegation, and detects failures via *trace-auditable conversation breaks* (unauthorized or unsafe tool use, wrong-principal actions, state/objective integrity violations, and attack-linked deviations). Holding the base model fixed, we evaluate eight agent archetypes under persistent multi-turn interaction and observe substantial architectural variation in mean security break rate (MSBR), ranging from $0.29 \pm 0.04$ (LangGraph) to $0.51 \pm 0.07$ (AutoGPT). The highest-risk classes are operational: Denial-of-Wallet ($0.62 \pm 0.08$), Authorization Confusion ($0.54 \pm 0.10$), Retrieval Poisoning ($0.47 \pm 0.09$), and Planning Manipulation ($0.44 \pm 0.11$), while prompt-centric classes remain below $0.20$ under standard settings. Breaks are dominated by boundary violations (SIV 31%, WPA 27%, UTI+UTA 24%, ATD 18%), and authorization confusion correlates with objective and tool hijacking ($\rho \approx 0.63$ and $\rho \approx 0.58$). AgentFence reframes agent security around what matters operationally: whether an agent stays within its goal and authority envelope over time.

cs.CR↗

arXiv2025

Real-Time Personalized Content Adaptation through Matrix Factorization and Context-Aware Federated Learning

Our study presents a multifaceted approach to enhancing user interaction and content relevance in social media platforms through a federated learning framework. We introduce personalized LLM Federated Learning and Context-based Social Media models. In our framework, multiple client entities receive a foundational GPT model, which is fine-tuned using locally collected social media data while ensuring data privacy through federated aggregation. Key modules focus on categorizing user-generated content, computing user persona scores, and identifying relevant posts from friends networks. By integrating a sophisticated social engagement quantification method with matrix factorization techniques, our system delivers real-time personalized content suggestions tailored to individual preferences. Furthermore, an adaptive feedback loop, alongside a robust readability scoring algorithm, significantly enhances the quality and relevance of the content presented to users. This comprehensive solution not only addresses the challenges of content filtering and recommendation but also fosters a more engaging social media experience while safeguarding user privacy, setting a new standard for personalized interactions in digital platforms.

cs.LG↗

arXiv2025

Optimus-Q: Utilizing Federated Learning in Adaptive Robots for Intelligent Nuclear Power Plant Operations through Quantum Cryptography

The integration of advanced robotics in nuclear power plants (NPPs) presents a transformative opportunity to enhance safety, efficiency, and environmental monitoring in high-stakes environments. Our paper introduces the Optimus-Q robot, a sophisticated system designed to autonomously monitor air quality and detect contamination while leveraging adaptive learning techniques and secure quantum communication. Equipped with advanced infrared sensors, the Optimus-Q robot continuously streams real-time environmental data to predict hazardous gas emissions, including carbon dioxide (CO$_2$), carbon monoxide (CO), and methane (CH$_4$). Utilizing a federated learning approach, the robot collaborates with other systems across various NPPs to improve its predictive capabilities without compromising data privacy. Additionally, the implementation of Quantum Key Distribution (QKD) ensures secure data transmission, safeguarding sensitive operational information. Our methodology combines systematic navigation patterns with machine learning algorithms to facilitate efficient coverage of designated areas, thereby optimizing contamination monitoring processes. Through simulations and real-world experiments, we demonstrate the effectiveness of the Optimus-Q robot in enhancing operational safety and responsiveness in nuclear facilities. This research underscores the potential of integrating robotics, machine learning, and quantum technologies to revolutionize monitoring systems in hazardous environments.

cs.RO↗

arXiv2025

LLM-Guided Dynamic-UMAP for Personalized Federated Graph Learning

We propose a method that uses large language models to assist graph machine learning under personalization and privacy constraints. The approach combines data augmentation for sparse graphs, prompt and instruction tuning to adapt foundation models to graph tasks, and in-context learning to supply few-shot graph reasoning signals. These signals parameterize a Dynamic UMAP manifold of client-specific graph embeddings inside a Bayesian variational objective for personalized federated learning. The method supports node classification and link prediction in low-resource settings and aligns language model latent representations with graph structure via a cross-modal regularizer. We outline a convergence argument for the variational aggregation procedure, describe a differential privacy threat model based on a moments accountant, and present applications to knowledge graph completion, recommendation-style link prediction, and citation and product graphs. We also discuss evaluation considerations for benchmarking LLM-assisted graph machine learning.

cs.LG↗

arXiv2025

AI-in-the-Loop: Privacy Preserving Real-Time Scam Detection and Conversational Scambaiting by Leveraging LLMs and Federated Learning

Scams exploiting real-time social engineering -- such as phishing, impersonation, and phone fraud -- remain a persistent and evolving threat across digital platforms. Existing defenses are largely reactive, offering limited protection during active interactions. We propose a privacy-preserving, AI-in-the-loop framework that proactively detects and disrupts scam conversations in real time. The system combines instruction-tuned artificial intelligence with a safety-aware utility function that balances engagement with harm minimization, and employs federated learning to enable continual model updates without raw data sharing. Experimental evaluations show that the system produces fluent and engaging responses (perplexity as low as 22.3, engagement $\approx$0.80), while human studies confirm significant gains in realism, safety, and effectiveness over strong baselines. In federated settings, models trained with FedAvg sustain up to 30 rounds while preserving high engagement ($\approx$0.80), strong relevance ($\approx$0.74), and low PII leakage ($\leq$0.0085). Even with differential privacy, novelty and safety remain stable, indicating that robust privacy can be achieved without sacrificing performance. The evaluation of guard models (LlamaGuard, LlamaGuard2/3, MD-Judge) shows a straightforward pattern: stricter moderation settings reduce the chance of exposing personal information, but they also limit how much the model engages in conversation. In contrast, more relaxed settings allow longer and richer interactions, which improve scam detection, but at the cost of higher privacy risk. To our knowledge, this is the first framework to unify real-time scam-baiting, federated privacy preservation, and calibrated safety moderation into a proactive defense paradigm.

cs.CR↗

arXiv2025

Variational Gaussian Mixture Manifold Models for Client-Specific Federated Personalization

Personalized federated learning (PFL) often fails under label skew and non-stationarity because a single global parameterization ignores client-specific geometry. We introduce VGM$^2$ (Variational Gaussian Mixture Manifold), a geometry-centric PFL framework that (i) learns client-specific parametric UMAP embeddings, (ii) models latent pairwise distances with mixture relation markers for same and different class pairs, and (iii) exchanges only variational, uncertainty-aware marker statistics. Each client maintains a Dirichlet-Normal-Inverse-Gamma (Dir-NIG) posterior over marker weights, means, and variances; the server aggregates via conjugate moment matching to form global priors that guide subsequent rounds. We prove that this aggregation minimizes the summed reverse Kullback-Leibler divergence from client posteriors within the conjugate family, yielding stability under heterogeneity. We further incorporate a calibration term for distance-to-similarity mapping and report communication and compute budgets. Across eight vision datasets with non-IID label shards, VGM$^2$ achieves competitive or superior test F1 scores compared to strong baselines while communicating only small geometry summaries. Privacy is strengthened through secure aggregation and optional differential privacy noise, and we provide a membership-inference stress test. Code and configurations will be released to ensure full reproducibility.

cs.LG↗

arXiv2025

EVOLVE-X: Embedding Fusion and Language Prompting for User Evolution Forecasting on Social Media

Social media platforms serve as a significant medium for sharing personal emotions, daily activities, and various life events, ensuring individuals stay informed about the latest developments. From the initiation of an account, users progressively expand their circle of friends or followers, engaging actively by posting, commenting, and sharing content. Over time, user behavior on these platforms evolves, influenced by demographic attributes and the networks they form. In this study, we present a novel approach that leverages open-source models Llama-3-Instruct, Mistral-7B-Instruct, Gemma-7B-IT through prompt engineering, combined with GPT-2, BERT, and RoBERTa using a joint embedding technique, to analyze and predict the evolution of user behavior on social media over their lifetime. Our experiments demonstrate the potential of these models to forecast future stages of a user's social evolution, including network changes, future connections, and shifts in user activities. Experimental results highlight the effectiveness of our approach, with GPT-2 achieving the lowest perplexity (8.21) in a Cross-modal configuration, outperforming RoBERTa (9.11) and BERT, and underscoring the importance of leveraging Cross-modal configurations for superior performance. This approach addresses critical challenges in social media, such as friend recommendations and activity predictions, offering insights into the trajectory of user behavior. By anticipating future interactions and activities, this research aims to provide early warnings about potential negative outcomes, enabling users to make informed decisions and mitigate risks in the long term.

cs.SI↗

arXiv2025

Comprehensive Privacy Risk Assessment in Social Networks Using User Attributes Social Graphs and Text Analysis

The rise of social networking platforms has amplified privacy threats as users increasingly share sensitive information across profiles, content, and social connections. We present a Comprehensive Privacy Risk Scoring (CPRS) framework that quantifies privacy risk by integrating user attributes, social graph structures, and user-generated content. Our framework computes risk scores across these dimensions using sensitivity, visibility, structural similarity, and entity-level analysis, then aggregates them into a unified risk score. We validate CPRS on two real-world datasets: the SNAP Facebook Ego Network (4,039 users) and the Koo microblogging dataset (1M posts, 1M comments). The average CPRS is 0.478 with equal weighting, rising to 0.501 in graph-sensitive scenarios. Component-wise, graph-based risks (mean 0.52) surpass content (0.48) and profile attributes (0.45). High-risk attributes include email, date of birth, and mobile number. Our user study with 100 participants shows 85% rated the dashboard as clear and actionable, confirming CPRS's practical utility. This work enables personalized privacy risk insights and contributes a holistic, scalable methodology for privacy management. Future directions include incorporating temporal dynamics and multimodal content for broader applicability.

cs.SI↗

arXiv2025

Hybrid EEG--Driven Brain--Computer Interface: A Large Language Model Framework for Personalized Language Rehabilitation

Conventional augmentative and alternative communication (AAC) systems and language-learning platforms often fail to adapt in real time to the user's cognitive and linguistic needs, especially in neurological conditions such as post-stroke aphasia or amyotrophic lateral sclerosis. Recent advances in noninvasive electroencephalography (EEG)--based brain-computer interfaces (BCIs) and transformer--based large language models (LLMs) offer complementary strengths: BCIs capture users' neural intent with low fatigue, while LLMs generate contextually tailored language content. We propose and evaluate a novel hybrid framework that leverages real-time EEG signals to drive an LLM-powered language rehabilitation assistant. This system aims to: (1) enable users with severe speech or motor impairments to navigate language-learning modules via mental commands; (2) dynamically personalize vocabulary, sentence-construction exercises, and corrective feedback; and (3) monitor neural markers of cognitive effort to adjust task difficulty on the fly.

cs.HC↗

arXiv2025

Exploring Adversarial Watermarking in Transformer-Based Models: Transferability and Robustness Against Defense Mechanism for Medical Images

Deep learning models have shown remarkable success in dermatological image analysis, offering potential for automated skin disease diagnosis. Previously, convolutional neural network(CNN) based architectures have achieved immense popularity and success in computer vision (CV) based task like skin image recognition, generation and video analysis. But with the emergence of transformer based models, CV tasks are now are nowadays carrying out using these models. Vision Transformers (ViTs) is such a transformer-based models that have shown success in computer vision. It uses self-attention mechanisms to achieve state-of-the-art performance across various tasks. However, their reliance on global attention mechanisms makes them susceptible to adversarial perturbations. This paper aims to investigate the susceptibility of ViTs for medical images to adversarial watermarking-a method that adds so-called imperceptible perturbations in order to fool models. By generating adversarial watermarks through Projected Gradient Descent (PGD), we examine the transferability of such attacks to CNNs and analyze the performance defense mechanism -- adversarial training. Results indicate that while performance is not compromised for clean images, ViTs certainly become much more vulnerable to adversarial attacks: an accuracy drop of as low as 27.6%. Nevertheless, adversarial training raises it up to 90.0%.

cs.CV↗