arXiv subjects

Jun Sakuma

Publications and source records attributed to Jun Sakuma.

At least 19 recordsLinked to original sources

arXiv2026

Capability-Routed Guard: Defending Large Reasoning Models Against Reasoning-Centric Jailbreaks

Large reasoning models (LRMs) expose a new safety failure mode: adversarial prompts can manipulate reasoning context, task decomposition, or capability interpretation so that harmful objectives are processed as legitimate reasoning steps. Existing safeguards, including safety reminders, external classifiers, and self-checking wrappers, are often brittle because they either inspect the adversarial prompt directly or ask the target model to perform additional safety reasoning on the same surface that attacks exploit. We introduce Capability-Routed Guard (CRG), a model-agnostic inference-time guardrail for closed-source LRMs, where defenders cannot inspect hidden reasoning traces or modify model weights. CRG reframes prompt defense as a capability-routing problem: a side-channel controller first constructs a trusted representation of the user's authorized task, active context, safety evidence, and capability-transfer risk, separating executable intent from untrusted reasoning context. This representation supports route-specific execution, allowing CRG to block high-risk requests, constrain ambiguous ones, and forward low-risk requests through trusted active context. Finally, CRG applies TraceCheck to verify consistency with the authorized task and invokes a restricted fallback to preserve utility for low-risk benign prompts. Extensive experiments demonstrate that CRG effectively mitigates diverse reasoning-centric jailbreaks while preserving benign utility and avoiding common over-refusal issues. Further analysis shows that its components contribute complementary benefits, highlighting the importance of coordinated defense mechanisms for securing large reasoning models.

cs.CR↗

arXiv2026

Harmless Yet Harmful: Neutral Prompting Attacks for Stealthy Hallucination Steering in Agent Skills

LLM-powered coding agents increasingly participate in software development workflows by generating code, selecting dependencies, and producing package installation commands. This creates a new software supply chain risk: when an agent hallucinates a non-existent package, an attacker may register the hallucinated name and later compromise users who install it. Existing package hallucination attacks and defenses primarily focus on naturally occurring hallucinations, targeted dependency steering, or post-hoc package validation. In this paper, we introduce \emph{Neutral Prompting Attack} (NPA), a highly stealthy attack paradigm in which semantically benign instructions, such as encouraging imagination and exhaustiveness, increase package hallucination propensity without containing explicit malicious intent. Unlike targeted dependency steering, NPA does not specify an attacker-chosen package. Instead, it shifts the model's dependency generation behavior toward more speculative package names. We evaluate NPA across multiple coding-oriented LLMs and package hallucination benchmarks. Our results show that NPA increases both \emph{Hallucination ASR} and \emph{Pip Install ASR}, changes the distribution of hallucinated package names, and evades existing static-analysis, LLM-based, and agent-based Skill defenses. These findings reveal that harmless-looking prompts can covertly manipulate hallucination behavior and create downstream software supply chain risks.

cs.CR↗

arXiv2026

MetaBackdoor: Exploiting Positional Encoding as a Backdoor Attack Surface in LLMs

Backdoor attacks pose a serious security threat to large language models (LLMs), which are increasingly deployed as general-purpose assistants in safety- and privacy-critical applications. Existing LLM backdoors rely primarily on content-based triggers, requiring explicit modification of the input text. In this work, we show that this assumption is unnecessary and limiting. We introduce MetaBackdoor, a new class of backdoor attacks that exploits positional information as the trigger, without modifying textual content. Our key insight is that Transformer-based LLMs necessarily encode token positions to process ordered sequences. As a result, length-correlated positional structure is reflected in the model's internal computation and can be used as an effective non-content trigger signal. We demonstrate that even a simple length-based positional trigger is sufficient to activate stealthy backdoors. Unlike prior attacks, MetaBackdoor operates on visibly and semantically clean inputs and enables qualitatively new capabilities. We show that a backdoored LLM can be induced to disclose sensitive internal information, including proprietary system prompts, once a length condition is satisfied. We further demonstrate a self-activation scenario, where normal multi-turn interaction can move the conversation context into the trigger region and induce malicious tool-call behavior without attacker-supplied trigger text. In addition, MetaBackdoor is orthogonal to content-based backdoors and can be composed with them to create more precise and harder-to-detect activation conditions. Our results expand the threat model of LLM backdoors by revealing positional encoding as a previously overlooked attack surface. This challenges defenses that focus on detecting suspicious text and highlights the need for new defense strategies that explicitly account for positional triggers in modern LLM architectures.

cs.CR↗

arXiv2026

In situ U Pb chronology and chemistry of zirconolite in the andesitic meteorite Erg Chech 002

Precise and accurate ages for asteroidal crusts are fundamental for reconstructing the timeline of magmatic, metamorphic, and impact events in the early Solar System. Zirconolite (CaZrTi2O7) is an accessory mineral found in a wide range of crustal rocks on both the Earth and Moon, and has proven to be a potentially useful U Pb chronometer. However, this mineral is rare in asteroidal meteorites, and its use for early Solar System chronology has been limited. We present the in situ occurrence, U Pb chronology, and chemistry of zirconolite in the andesitic meteorite Erg Chech 002, which represents a sample of the oldest known asteroidal crust. The zirconolite occurs as needle- and fiber-shaped, stubby crystals, with widths of 3 micrometers and lengths of up to 30 micrometers. Electron and ion microprobe analyses yielded concordant U Pb data with a weighted mean 207Pb 206Pb age of 4557.9 4.3 Ma 2sigma, rendering it the Solar System's oldest known zirconolite. Yet, this age is distinctly younger than reported high-precision 207Pb 206Pb ages varying from 4565.6 to 4566.2 Ma, which were obtained by acid leaching of pyroxene and whole rock samples of the meteorite. From its mineralogical and REE (U+Th) (Nb+Ta) characteristics, we argue that the zirconolite age represents the timing of a shock metamorphism of the parent asteroid's crust. Our results suggest that 207Pb 206Pb dating for acid-leached samples can be affected by including even a tiny amount of metamorphic zirconolite, calling for caution in interpreting the high-precision 207Pb 206Pb age data. On the basis of thermodynamic and geochemical considerations, we infer further occurrences of zirconolite in alkali-silica-rich asteroidal rocks that rapidly cooled from high temperatures.

physics.geo-ph↗

arXiv2026

When Benchmarks Leak: Inference-Time Decontamination for LLMs

Benchmark-based evaluation is the de facto standard for comparing large language models (LLMs). However, its reliability is increasingly threatened by test set contamination, where test samples or their close variants leak into training data and artificially inflate reported performance. To address this issue, prior work has explored two main lines of mitigation. One line attempts to identify and remove contaminated benchmark items before evaluation, but this inevitably alters the evaluation set itself and becomes unreliable when contamination is moderate or severe. The other line preserves the benchmark and instead suppresses contaminated behavior at evaluation time; however, such interventions often interfere with normal inference and lead to noticeable performance degradation on clean inputs. We propose DeconIEP, a decontamination framework that operates entirely during evaluation by applying small, bounded perturbations in the input embedding space. Guided by a relatively less-contaminated reference model, DeconIEP learns an instance-adaptive perturbation generator that steers the evaluated model away from memorization-driven shortcut pathways. Across multiple open-weight LLMs and benchmarks, extensive empirical results show that DeconIEP achieves strong decontamination effectiveness while incurring only minimal degradation in benign utility.

cs.CL↗

arXiv2025

One Leak Away: How Pretrained Model Exposure Amplifies Jailbreak Risks in Finetuned LLMs

Finetuning pretrained large language models (LLMs) has become the standard paradigm for developing downstream applications. However, its security implications remain unclear, particularly regarding whether finetuned LLMs inherit jailbreak vulnerabilities from their pretrained sources. We investigate this question in a realistic pretrain-to-finetune threat model, where an attacker has full access to a released pretrained LLM but no access to its proprietary finetuned derivatives. Empirical analysis shows that adversarial prompts optimized on the pretrained model transfer most effectively to its finetuned variants, revealing inherited vulnerabilities from pretrained to finetuned LLMs. To further examine this inheritance, we conduct representation-level probing, which shows that transferable prompts are linearly separable within the pretrained hidden states, suggesting that transferability-relevant structure is already encoded in pretrained representations. Building on this insight, we propose the Probe-Guided Projection (PGP) attack, which steers optimization toward transferability-relevant directions. Experiments across multiple LLM families and diverse finetuned tasks confirm PGP's strong transfer success, underscoring the security risks inherent in the pretrain-to-finetune paradigm. Finally, we demonstrate that the same representation-level insights also enable a lightweight defense that mitigates pretrain-to-finetune jailbreak transfer while preserving downstream utility.

cs.CR↗

arXiv2025

Model-agnostic Adversarial Attack and Defense for Vision-Language-Action Models

Vision-Language-Action (VLA) models have achieved revolutionary progress in robot learning, enabling robots to execute complex physical robot tasks from natural language instructions. Despite this progress, their adversarial robustness remains underexplored. In this work, we propose both adversarial patch attack and corresponding defense strategies for VLA models. We first introduce the Embedding Disruption Patch Attack (EDPA), a model-agnostic adversarial attack that generates patches directly placeable within the camera's view. In comparison to prior methods, EDPA can be readily applied to different VLA models without requiring prior knowledge of the model architecture, or the controlled robotic manipulator. EDPA constructs these patches by (i) disrupting the semantic alignment between visual and textual latent representations, and (ii) maximizing the discrepancy of latent representations between adversarial and corresponding clean visual inputs. Through the optimization of these objectives, EDPA distorts the VLA's interpretation of visual information, causing the model to repeatedly generate incorrect actions and ultimately result in failure to complete the given robotic task. To counter this, we propose an adversarial fine-tuning scheme for the visual encoder, in which the encoder is optimized to produce similar latent representations for both clean and adversarially perturbed visual inputs. Extensive evaluations on the widely recognized LIBERO robotic simulation benchmark demonstrate that EDPA substantially increases the task failure rate of cutting-edge VLA models, while our proposed defense effectively mitigates this degradation. The codebase is accessible via the homepage at https://edpa-attack.github.io/.

cs.CV↗

arXiv2025

Pattern Enhanced Multi-Turn Jailbreaking: Exploiting Structural Vulnerabilities in Large Language Models

Large language models (LLMs) remain vulnerable to multi-turn jailbreaking attacks that exploit conversational context to bypass safety constraints gradually. These attacks target different harm categories through distinct conversational approaches. Existing multi-turn methods often rely on heuristic or ad hoc exploration strategies, providing limited insight into underlying model weaknesses. The relationship between conversation patterns and model vulnerabilities across harm categories remains poorly understood. We propose Pattern Enhanced Chain of Attack (PE-CoA), a framework of five conversation patterns to construct multi-turn jailbreaks through natural dialogue. Evaluating PE-CoA on twelve LLMs spanning ten harm categories, we achieve state-of-the-art performance, uncovering pattern-specific vulnerabilities and LLM behavioral characteristics: models exhibit distinct weakness profiles, defense to one pattern does not generalize to others, and model families share similar failure modes. These findings highlight limitations of safety training and indicate the need for pattern-aware defenses. Code available on: https://github.com/Ragib-Amin-Nihal/PE-CoA

cs.CL↗

arXiv2025

Differentially Private Synthetic Text Generation for Retrieval-Augmented Generation (RAG)

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by grounding them in external knowledge. However, its application in sensitive domains is limited by privacy risks. Existing private RAG methods typically rely on query-time differential privacy (DP), which requires repeated noise injection and leads to accumulated privacy loss. To address this issue, we propose DP-SynRAG, a framework that uses LLMs to generate differentially private synthetic RAG databases. Unlike prior methods, the synthetic text can be reused once created, thereby avoiding repeated noise injection and additional privacy costs. To preserve essential information for downstream RAG tasks, DP-SynRAG extends private prediction, which instructs LLMs to generate text that mimics subsampled database records in a DP manner. Experiments show that DP-SynRAG achieves superior performance to the state-of-the-art private RAG systems while maintaining a fixed privacy budget, offering a scalable solution for privacy-preserving RAG.

cs.CR↗

arXiv2025

Toward Safer Diffusion Language Models: Discovery and Mitigation of Priming Vulnerability

Diffusion language models (DLMs) generate tokens in parallel through iterative denoising, which can reduce latency and enable bidirectional conditioning. However, the safety risks posed by jailbreak attacks that exploit this inference mechanism are not well understood. In this paper, we reveal that DLMs have a critical vulnerability stemming from their iterative denoising process and propose a countermeasure. Specifically, our investigation shows that if an affirmative token for a harmful query appears at an intermediate step, subsequent denoising can be steered toward a harmful response even in aligned models. As a result, simply injecting such affirmative tokens can readily bypass the safety guardrails. Furthermore, we demonstrate that the vulnerability allows existing optimization-based jailbreak attacks to succeed on DLMs. Building on this analysis, we propose a novel safety alignment method tailored to DLMs that trains models to generate safe responses from contaminated intermediate states that contain affirmative tokens. Our experiments indicate that the proposed method significantly mitigates the vulnerability with minimal impact on task performance. Furthermore, our method improves robustness against conventional jailbreak attacks. Our work underscores the need for DLM-specific safety research. Our code is available at https://github.com/mdl-lab/dlm-priming-vulnerability.

cs.AI↗

arXiv2025

Rewriting Pre-Training Data Boosts LLM Performance in Math and Code

The performance of large language models (LLMs) in program synthesis and mathematical reasoning is fundamentally limited by the quality of their pre-training corpora. We introduce two openly licensed pre-training datasets, released under the Llama 3.3 Community License, that significantly enhance LLM performance by systematically rewriting public data. SwallowCode ($\approx$16.1 billion tokens) refines Python snippets from The-Stack-v2 through a novel four-stage pipeline: syntax validation, pylint-based style filtering, and a two-stage LLM rewriting process that enforces style conformity and transforms snippets into self-contained, algorithmically efficient examples. Unlike prior methods that rely on exclusionary filtering or limited transformations, our transform-and-retain approach refines low-quality code, maximizing data utility. SwallowMath ($\approx$2.3 billion tokens) enhances Finemath-4+ by removing boilerplate, restoring context, and reformatting solutions into concise, step-by-step explanations. Within a fixed 50 billion token training budget, continual pre-training of Llama-3.1-8B with SwallowCode boosts pass@1 by +17.0 on HumanEval and +16.1 on HumanEval+ compared to Stack-Edu, surpassing the baseline model's code generation capabilities. Similarly, substituting SwallowMath yields +12.4 accuracy on GSM8K and +7.6 on MATH. Ablation studies confirm that each pipeline stage contributes incrementally, with rewriting yielding the largest gains. By releasing datasets, prompts, checkpoints, and pipeline code, we ensure reproducibility and provide a transferable transform-and-retain methodology that can be adapted to other base models and LLM rewriting setups.

cs.LG↗

arXiv2025

Data Poisoning Attacks to Locally Differentially Private Range Query Protocols

Local Differential Privacy (LDP) has been widely adopted to protect user privacy in decentralized data collection. However, recent studies have revealed that LDP protocols are vulnerable to data poisoning attacks, where malicious users manipulate their reported data to distort aggregated results. In this work, we present the first study on data poisoning attacks targeting LDP range query protocols, focusing on both tree-based and grid-based approaches. We identify three key challenges in executing such attacks, including crafting consistent and effective fake data, maintaining data consistency across levels or grids, and preventing server detection. To address the first two challenges, we propose novel attack methods that are provably optimal, including a tree-based attack and a grid-based attack, designed to manipulate range query results with high effectiveness. \textbf{Our key finding is that the common post-processing procedure, Norm-Sub, in LDP range query protocols can help the attacker massively amplify their attack effectiveness.} In addition, we study a potential countermeasure, but also propose an adaptive attack capable of evading this defense to address the third challenge. We evaluate our methods through theoretical analysis and extensive experiments on synthetic and real-world datasets. Our results show that the proposed attacks can significantly amplify estimations for arbitrary range queries by manipulating a small fraction of users, providing 5-10x more influence than a normal user to the estimation.

cs.CR↗

arXiv2025

Explainable Classifier for Malignant Lymphoma Subtyping via Cell Graph and Image Fusion

Malignant lymphoma subtype classification directly impacts treatment strategies and patient outcomes, necessitating classification models that achieve both high accuracy and sufficient explainability. This study proposes a novel explainable Multi-Instance Learning (MIL) framework that identifies subtype-specific Regions of Interest (ROIs) from Whole Slide Images (WSIs) while integrating cell distribution characteristics and image information. Our framework simultaneously addresses three objectives: (1) indicating appropriate ROIs for each subtype, (2) explaining the frequency and spatial distribution of characteristic cell types, and (3) achieving high-accuracy subtyping by leveraging both image and cell-distribution modalities. The proposed method fuses cell graph and image features extracted from each patch in the WSI using a Mixture-of-Experts (MoE) approach and classifies subtypes within an MIL framework. Experiments on a dataset of 1,233 WSIs demonstrate that our approach achieves state-of-the-art accuracy among ten comparative methods and provides region-level and cell-level explanations that align with a pathologist's perspectives.

cs.CV↗

arXiv2025

BADTV: Unveiling Backdoor Threats in Third-Party Task Vectors

Task arithmetic in large-scale pre-trained models enables agile adaptation to diverse downstream tasks without extensive retraining. By leveraging task vectors (TVs), users can perform modular updates through simple arithmetic operations like addition and subtraction. Yet, this flexibility presents new security challenges. In this paper, we investigate how TVs are vulnerable to backdoor attacks, revealing how malicious actors can exploit them to compromise model integrity. By creating composite backdoors that are designed asymmetrically, we introduce BadTV, a backdoor attack specifically crafted to remain effective simultaneously under task learning, forgetting, and analogy operations. Extensive experiments show that BadTV achieves near-perfect attack success rates across diverse scenarios, posing a serious threat to models relying on task arithmetic. We also evaluate current defenses, finding they fail to detect or mitigate BadTV. Our results highlight the urgent need for robust countermeasures to secure TVs in real-world deployments.

cs.LG↗

arXiv2024

Beyond Full Poisoning: Effective Availability Attacks with Partial Perturbation

The widespread use of publicly available datasets for training machine learning models raises significant concerns about data misuse. Availability attacks have emerged as a means for data owners to safeguard their data by designing imperceptible perturbations that degrade model performance when incorporated into training datasets. However, existing availability attacks are ineffective when only a portion of the data can be perturbed. To address this challenge, we propose a novel availability attack approach termed Parameter Matching Attack (PMA). PMA is the first availability attack capable of causing more than a 30\% performance drop when only a portion of data can be perturbed. PMA optimizes perturbations so that when the model is trained on a mixture of clean and perturbed data, the resulting model will approach a model designed to perform poorly. Experimental results across four datasets demonstrate that PMA outperforms existing methods, achieving significant model performance degradation when a part of the training data is perturbed. Our code is available in the supplementary materials.

cs.LG↗

arXiv2024

Zero-shot domain adaptation based on dual-level mix and contrast

Zero-shot domain adaptation (ZSDA) is a domain adaptation problem in the situation that labeled samples for a target task (task of interest) are only available from the source domain at training time, but for a task different from the task of interest (irrelevant task), labeled samples are available from both source and target domains. In this situation, classical domain adaptation techniques can only learn domain-invariant features in the irrelevant task. However, due to the difference in sample distribution between the two tasks, domain-invariant features learned in the irrelevant task are biased and not necessarily domain-invariant in the task of interest. To solve this problem, this paper proposes a new ZSDA method to learn domain-invariant features with low task bias. To this end, we propose (1) data augmentation with dual-level mixups in both task and domain to fill the absence of target task-of-interest data, (2) an extension of domain adversarial learning to learn domain-invariant features with less task bias, and (3) a new dual-level contrastive learning method that enhances domain-invariance and less task biasedness of features. Experimental results show that our proposal achieves good performance on several benchmarks.

cs.CV↗

arXiv2024

Robust Deep Reinforcement Learning against Adversarial Behavior Manipulation

This study investigates behavior-targeted attacks on reinforcement learning and their countermeasures. Behavior-targeted attacks aim to manipulate the victim's behavior as desired by the adversary through adversarial interventions in state observations. Existing behavior-targeted attacks have some limitations, such as requiring white-box access to the victim's policy. To address this, we propose a novel attack method using imitation learning from adversarial demonstrations, which works under limited access to the victim's policy and is environment-agnostic. In addition, our theoretical analysis proves that the policy's sensitivity to state changes impacts defense performance, particularly in the early stages of the trajectory. Based on this insight, we propose time-discounted regularization, which enhances robustness against attacks while maintaining task performance. To the best of our knowledge, this is the first defense strategy specifically designed for behavior-targeted attacks.

cs.LG↗

arXiv2024

Harnessing the Power of Vicinity-Informed Analysis for Classification under Covariate Shift

Transfer learning enhances prediction accuracy on a target distribution by leveraging data from a source distribution, demonstrating significant benefits in various applications. This paper introduces a novel dissimilarity measure that utilizes vicinity information, i.e., the local structure of data points, to analyze the excess error in classification under covariate shift, a transfer learning setting where marginal feature distributions differ but conditional label distributions remain the same. We characterize the excess error using the proposed measure and demonstrate faster or competitive convergence rates compared to previous techniques. Notably, our approach is effective in the support non-containment assumption, which often appears in real-world applications, holds. Our theoretical analysis bridges the gap between current theoretical findings and empirical observations in transfer learning, particularly in scenarios with significant differences between source and target distributions.

stat.ML↗