Searcharxiv⌕ Search

arXiv subjects

Wanlei Zhou

Publications and source records attributed to Wanlei Zhou.

At least 19 recordsLinked to original sources

Hijacking a Trusted LLM in Real-World Quantized Deployments Only Requires a Disguised Adapter

Third-party low-rank adaptation (LoRA) adapters are tailored to perform decision tasks such as content moderation and function selection. However, in the real world, quantized deployments carry a potential threat in that an adapter can pass an audit with the unquantized base, yet hijack decisions when the same base is quantized for deployment. Demonstrating this threat requires an adapter that preserves benign behavior before quantization, but hijacks decisions afterward. Further, this must hold without modifying the trusted base or knowing the victim's quantization configuration. We therefore present QHijack-LoRA, an adapter-only attack that uses the base model's natural quantization residual to activate decision hijacking. The attacker modifies neither the trusted base model nor the victim's quantization configuration. With QHijack-LoRA, a four-objective training procedure separates full-precision dormancy from quantized target activation, while preserving clean-input behavior. The adapter is trained across several different quantization methods, while adaptive reweighting and residual-direction augmentation extend the attack to unseen configurations. Our results show that auditing performed exclusively under full-precision conditions does not provide assurance of decision integrity in quantized deployment environments. Thus, we also propose a cross-precision paired audit that checks the candidate adapter in its intended deployment configuration. This work shows the critical need to audit third-party adapters in any potential quantized deployment.

cs.CR↗

Protecting Sensitive Knowledge Under LLM Unlearning: The Proxy Threat and a Superposition-Aware Defense

Machine unlearning lets the owner of large language models (LLMs) remove specific entities from a trained model without retraining from scratch. Existing machine unlearning methods focus on removing the requested entities while maintaining utility. However, we observe that the representations of distinct entities in an LLM may overlap in shared parameter subspaces, a phenomenon known as knowledge superposition. Removing one entity may then affect other entities. This exposes a new security threat: legitimate unlearning requests can be exploited to remove sensitive entities that were never requested. This threat is unlike data poisoning or backdoor attacks. We define the attack as the Proxy Unlearning Attack (PUA), which selects removable proxy entities superposed with a sensitive target. The attack can even occur unintentionally, when benign users accidentally remove entities superposed with a sensitive one. In response, we propose Superposition-Aware Unlearning Protection (SAUP), a general defense framework that can be applied to unlearning requests to ensure a correct removal. SAUP performs in two stages. Before unlearning, it uses superposition detection to estimate which sensitive entities the request puts at risk. After unlearning, it decouples the shared representations and restores the affected entities, while keeping the requested entities suppressed. Our findings reveal that legitimate unlearning requests can be maliciously exploited through knowledge superposition. When dealing with unlearning requests, LLM providers need representation-level detection and recovery mechanisms to mitigate this threat and protect sensitive knowledge.

cs.CR↗

Class Machine Unlearning for Complex Data via Concepts Inference and Data Poisoning

Machine unlearning aims to remove the influence of specified training data or knowledge from a trained model without requiring full retraining. This capability is particularly important for modern image classifiers and large language models (LLMs), where retraining can be computationally expensive. However, machine unlearning on complex data remains difficult because the target information is often distributed across multiple semantic elements. Existing methods mainly remove samples, modify labels, or edit model parameters to reduce the influence of the forgetting target. These approaches usually do not explicitly identify which semantic concepts connect the forgetting target to the model's prediction or generated response. As a result, it is difficult to determine which information to modify. This uncertainty may leave residual target information or unnecessarily affect knowledge that should be retained. To address this gap, we propose a concept-guided poisoning unlearning framework that explicitly identifies the concepts that contribute most strongly to the target class or knowledge and uses them to guide the model update. For image classification, our method first identifies class-relevant concepts with a Post-hoc Concept Bottleneck Model, localizes the image regions that express these concepts, and constructs replacement-based poisoned samples. For LLMs, it elicits the target knowledge through multiple questions, aggregates Integrated Gradients across the resulting responses to identify consistently important content, and masks this content to construct poisoned training targets. Experiments on multiple datasets show that the proposed framework achieves effective unlearning across different tasks while largely preserving retained model utility.

cs.LG↗

RISA: Response Inspection and Selective Actions for Refusal Calibration in Large Language Models

Reliable refusal behavior requires Large Language Models (LLMs) to reject harmful prompts with only answering benign ones. Incorrect refusal behavior can either expose users to harmful responses or prevent users from obtaining useful answers. Training-time alignment improves refusal behavior by updating model parameters with safety data, but requires additional computation and training. In contrast, inference-time alignment aims to modify LLM behavior during inference without updating the underlying model parameters. Existing inference-time methods mainly rely on in-context safety prompting, activation steering, or decoding control. However, most of them intervene without first determining whether the initial response is already appropriate, potentially altering a correct refusal or a useful answer. Effective selective intervention therefore requires identifying prompt intent beyond sensitive keywords, covering semantic variations that fixed rules may miss, and adapting the verifier to different base models. To address these challenges, we propose Response Inspection and Selective Actions (RISA), an inference-time framework that inspects the initial response and selectively corrects refusal errors without updating the base model. RISA first uses fixed contextual rules to assign refusal scores to clear cases. For unmatched cases, it derives a refusal score from the final-layer prompt hidden state using a calibrated linear probe. To adapt to different base models, RISA separately calibrates the probe score, representation-support boundary, and action thresholds. At runtime, RISA combines the prompt score with the initial refusal status and applies an action policy to intervene only when necessary. Experimental results demonstrate that RISA improves refusal reliability while largely preserving model utility, offering a practical solution for response-aware refusal calibration in LLMs.

cs.CR↗

BASIS: Breach-Aware Selective Prompt Injection Shielding with Prefill Attention Probes

Prompt injection is a critical security threat in large language model (LLM) applications, where attackers hijack model behavior by embedding malicious instructions in user or external data. Existing detection methods only detect the presence of injection and refuse to respond upon detection, overlooking the fact that for many modern aligned models, well-crafted instructions can resist most injection attacks. This means that the injection robustness varies significantly across instructions and models. This leads to widespread unnecessary over-refusal: inputs containing injections that the model could have handled correctly are rejected incorrectly. To deal with this over-refusal issue, we propose BASIS (Robustness-Aware Prompt Injection Defense). This defense method uses the Attention Competition Ratio ($ρ$) as features to train two sparse linear probes: an existence probe and a breach probe. Both probes make defense decisions through cascaded gating, which does not require additional LLM inference. BASIS comprises three stages: injection existence detection, per-sample breach prediction, and instruction robustness assessment; the online cascade refuses only when the model would actually be compromised and thus avoids over-refusal on robust instructions. Experiments across four tasks and six open-source LLMs show that BASIS maintains near-perfect injection detection while substantially reducing over-refusal on safe attack samples, especially under robust instruction templates.

cs.CR↗

When Machine Unlearning Meets Retrieval-Augmented Generation (RAG): Keep Secret or Forget Knowledge?

The deployment of large language models (LLMs) like ChatGPT and Gemini has shown their powerful natural language generation capabilities. However, these models can inadvertently learn and retain sensitive information and harmful content during training, raising significant ethical and legal concerns. To address these issues, machine unlearning has been introduced as a potential solution. While existing unlearning methods take into account the specific characteristics of LLMs, they often suffer from high computational demands, limited applicability, or the risk of catastrophic forgetting. To address these limitations, we propose a lightweight behavioral unlearning framework based on Retrieval-Augmented Generation (RAG) technology. By modifying the external knowledge base of RAG, we simulate the effects of forgetting without directly interacting with the unlearned LLM. We approach the construction of unlearned knowledge as a constrained optimization problem, deriving two key components that underpin the effectiveness of RAG-based unlearning. This RAG-based approach is particularly effective for closed-source LLMs, where existing unlearning methods often fail. We evaluate our framework through extensive experiments on both open-source and closed-source models, including ChatGPT, Gemini, Llama-2-7b-chat, and PaLM 2. The results demonstrate that our approach meets five key unlearning criteria: effectiveness, universality, harmlessness, simplicity, and robustness. Meanwhile, this approach can extend to multimodal large language models and LLM-based agents.

cs.CR↗

One Framework for All: Cross-Modal Membership Inference for Generative Models

Large generative models across text-to-text, text-to-image, and image-to-text modalities have been shown to pose significant privacy risks. One fundamental threat is membership inference attacks (MIA), which aim to determine whether a given data point was used in a model's training set. Although prior work has investigated MIAs against these three classes of generative models, existing approaches treat them in isolation and are not cross-applicable, thereby limiting their real-world utility. To address this limitation, we present the first comprehensive study of a unified membership inference framework that applies across text-to-text, text-to-image, and image-to-text modalities. Our approach is grounded in a key modality-agnostic observation: the output distribution of a generative model can approximate its training data distribution. Leveraging this property, we model the distributions of model-generated outputs and auxiliary non-member samples in a shared embedding space, and perform membership inference via likelihood ratio testing. We conduct extensive experiments in a strict black-box setting under both partial-knowledge and zero-knowledge threat models, and evaluate membership inference against both fine-tuning and pre-training data. Experimental results demonstrate our approach's superior performance in comparison to existing state-of-the-art methods, which are typically optimized for a single model class.

cs.LG↗

Auditing Machine Unlearning: A Systematic Research on Whether Models Truly Forget

Machine unlearning has been extensively studied in response to growing privacy concerns and regulatory requirements. However, auditing whether unlearning algorithms have truly erased the influence of specific data remains an open challenge. The lack of reliable and practical auditing mechanisms can lead to critical privacy risks, such as residual information leakage. This paper initiates a systematic investigation into whether existing unlearning algorithms can truly forget the designated data. We propose the first practical and general-purpose auditing framework for machine unlearning, inspired by the concept of proof of ignorance. Our framework addresses the key practicality limitations of existing methods by eliminating the need for retraining-from-scratch baselines, avoiding the training of large numbers of shadow models, and requiring no intrusive intervention in the original training process. To evaluate the effectiveness of our framework, we first conduct validation experiments to verify its soundness and completeness. We then perform comprehensive experiments across six datasets and ten representative unlearning methods. The results demonstrate that our framework reliably distinguishes between successful and failed unlearning. In particular, we observe that retraining-based and fine-tuning-based methods can achieve effective unlearning, even when the target data remain in the original dataset. In contrast, de-optimization-based methods fail to achieve true unlearning and instead degrade the model's performance. Fisher/Hessian-based methods also fail to unlearn requested data, even formal certification is provided. Moreover, we show that our framework is robust against fake unlearning attempts and generalizes well to large language models.

cs.LG↗

From Spark to Fire: Modeling and Mitigating Error Cascades in LLM-Based Multi-Agent Collaboration

Large Language Model-based Multi-Agent Systems (LLM-MAS) are increasingly applied to complex collaborative scenarios. However, their collaborative mechanisms may cause minor inaccuracies to gradually solidify into system-level false consensus through iteration. Such risks are difficult to trace since errors can propagate and amplify through message dependencies. Existing protections often rely on single-agent validation or require modifications to the collaboration architecture, which can weaken effective information flow and may not align with natural collaboration processes in real tasks. To address this, we propose a propagation dynamics model tailored for LLM-MAS that abstracts collaboration as a directed dependency graph and provides an early-stage risk criterion to characterize amplification risk. Through experiments on six mainstream frameworks, we identify three vulnerability classes: cascade amplification, topological sensitivity, and consensus inertia. We further instantiate an attack where injecting just a single atomic error seed leads to widespread failure. In response, we introduce a genealogy-graph-based governance layer, implemented as a message-layer plugin, that suppresses both endogenous and exogenous error amplification without altering the collaboration architecture. Experiments show that this approach prevents final infection in at least 89% of runs across operating modes and significantly mitigates the cascading spread of minor errors.

cs.MA↗

When LLMs Team Up: A Coordinated Attack Framework for Automated Cyber Intrusions

Automated intrusion-style workflows require LLM agents to reason over partial observations, tool outputs, and executable artifacts under bounded budgets. A single LLM instance often compresses evidence extraction, planning, execution, and validation into one context, which increases the risk of context drift and error propagation. Existing LLM-based multi-agent systems support general collaboration, but they do not explicitly model the role boundaries, artifact provenance, and cost constraints that characterize multi-stage intrusion workflows. This paper presents CAESAR, a coordinated multi-agent framework for controlled analysis of LLM-agent behavior in intrusion-style tasks. CAESAR decomposes the workflow into five typed roles and coordinates them through a bounded round protocol with a persistent knowledge base, a per-round workspace, validator-gated knowledge promotion, and capability-token write isolation. We evaluate CAESAR on 25 CTF tasks across five categories and four LLM backends. Compared with a single-agent baseline under matched budgets and tool access, CAESAR improves task success and reduces performance variance, with larger gains on tasks requiring multi-step exploit composition. A secondary simulated interactional-security study suggests that the role structure can transfer beyond code-native surfaces. The results indicate that role transitions, artifact provenance, and knowledge-promotion events provide useful structural signals for monitoring coordinated LLM-agent behavior beyond individual prompt and output inspection. The dataset, implementation, and evaluation logs are released at https://github.com/Xu-Qiu/CMAS.

cs.CR↗

Machine Unlearning on Pre-trained Models by Residual Feature Alignment Using LoRA

Machine unlearning is an emerging technology that removes a subset of the training data from a trained model without significantly affecting the model performance on the remaining data. This topic is becoming increasingly important in protecting user privacy and eliminating harmful or outdated data. The key challenge lies in effectively and efficiently unlearning specific information without compromising the model's utility on the retained data. For pre-trained models, fine-tuning is an important way to achieve the unlearning target. Previous work typically fine-tuned the entire model's parameters, which incurred significant computational costs. In addition, the fine-tuning process may cause shifts in the intermediate layer features, affecting the model's overall utility. In this work, we propose a novel and efficient machine unlearning method for pre-trained models. We term the method Residual Feature Alignment Unlearning. Specifically, we leverage LoRA (Low-Rank Adaptation) to decompose the model's intermediate features into pre-trained features and residual features. By adjusting the residual features, we align the unlearned model with the pre-trained model at the intermediate feature level to achieve both unlearning and remaining targets. The method aims to learn zero residuals on the retained set and shifted residuals on the unlearning set. Extensive experiments on numerous datasets validate the effectiveness of our approach.

cs.LG↗

CSC: Turning the Adversary's Poison against Itself

Poisoning-based backdoor attacks pose significant threats to deep neural networks by embedding triggers in training data, causing models to misclassify triggered inputs as adversary-specified labels while maintaining performance on clean data. Existing poison restraint-based defenses often suffer from inadequate detection against specific attack variants and compromise model utility through unlearning methods that lead to accuracy degradation. This paper conducts a comprehensive analysis of backdoor attack dynamics during model training, revealing that poisoned samples form isolated clusters in latent space early on, with triggers acting as dominant features distinct from benign ones. Leveraging these insights, we propose Cluster Segregation Concealment (CSC), a novel poison suppression defense. CSC first trains a deep neural network via standard supervised learning while segregating poisoned samples through feature extraction from early epochs, DBSCAN clustering, and identification of anomalous clusters based on class diversity and density metrics. In the concealment stage, identified poisoned samples are relabeled to a virtual class, and the model's classifier is fine-tuned using cross-entropy loss to replace the backdoor association with a benign virtual linkage, preserving overall accuracy. CSC was evaluated on four benchmark datasets against twelve poisoning-based attacks, CSC outperforms nine state-of-the-art defenses by reducing average attack success rates to near zero with minimal clean accuracy loss. Contributions include robust backdoor patterns identification, an effective concealment mechanism, and superior empirical validation, advancing trustworthy artificial intelligence.

cs.CR↗

Unreal Thinking: Chain-of-Thought Hijacking via Two-stage Backdoor

Large Language Models (LLMs) are increasingly deployed in settings where Chain-of-Thought (CoT) is interpreted by users. This creates a new safety risk: attackers may manipulate the model's observable CoT to make malicious behaviors. In open-weight ecosystems, such manipulation can be embedded in lightweight adapters that are easy to distribute and attach to base models. In practice, persistent CoT hijacking faces three main challenges: the difficulty of directly hijacking CoT tokens within one continuous long CoT-output sequence while maintaining stable downstream outputs, the scarcity of malicious CoT data, and the instability of naive backdoor injection methods. To address the data scarcity issue, we propose Multiple Reverse Tree Search (MRTS), a reverse synthesis procedure that constructs output-aligned CoTs from prompt-output pairs without directly eliciting malicious CoTs from aligned models. Building on MRTS, we introduce Two-stage Backdoor Hijacking (TSBH), which first induces a trigger-conditioned mismatch between intermediate CoT and malicious outputs, and then fine-tunes the model on MRTS-generated CoTs that have lower embedding distance to the malicious outputs, thereby ensuring stronger semantic similarity. Experiments across multiple open-weight models demonstrate that our method successfully induces trigger-activated CoT hijacking while maintaining a quantifiable distinction between hijacked and baseline states under our evaluation framework. We further explore a reasoning-based mitigation approach and release a safety-reasoning dataset to support future research on safety-aware and reliable reasoning. Our code is available at https://github.com/ChangWenhan/TSBH_official.

cs.CR↗

Are LLMs Ready for Computer Science Education? A Cross-Domain, Cross-Lingual and Cognitive-Level Evaluation Using Professional Certification Exams

Large language models (LLMs) are increasingly applied in computer science education for tasks such as tutoring, content generation, and code assessment. However, systematic evaluations aligned with formal curricula and certification standards remain limited. This study benchmarked four recent models, including GPT-5, DeepSeek-R1, Qwen-Plus, and Llama-3.3-70B-Instruct, using a dataset of 1,068 questions derived from six certification exams covering networking, office applications, and Java programming. We evaluated performance across language (Chinese vs. English), cognitive levels based on Bloom's Taxonomy, domain knowledge, confidence-accuracy alignment, and robustness to input masking. Results showed that GPT-5 performed best on English-language certifications, while Qwen-Plus performed better in Chinese contexts. DeepSeek-R1 achieved the most balanced cross-lingual performance, whereas Llama-3.3 showed clear limitations in higher-order reasoning and robustness. All models performed worse on more complex tasks. These findings provide empirical support for the integration of LLMs into computer science education and offer practical implications for curriculum design and assessment.

cs.CY↗

Secure Forgetting: A Framework for Privacy-Driven Unlearning in Large Language Model (LLM)-Based Agents

Large language model (LLM)-based agents have recently gained considerable attention due to the powerful reasoning capabilities of LLMs. Existing research predominantly focuses on enhancing the task performance of these agents in diverse scenarios. However, as LLM-based agents become increasingly integrated into real-world applications, significant concerns emerge regarding their accumulation of sensitive or outdated knowledge. Addressing these concerns requires the development of mechanisms that allow agents to selectively forget previously learned knowledge, giving rise to a new term LLM-based agent unlearning. This paper initiates research on unlearning in LLM-based agents. Specifically, we propose a novel and comprehensive framework that categorizes unlearning scenarios into three contexts: state unlearning (forgetting specific states or items), trajectory unlearning (forgetting sequences of actions) and environment unlearning (forgetting entire environments or categories of tasks). Within this framework, we introduce a natural language-based unlearning method that trains a conversion model to transform high-level unlearning requests into actionable unlearning prompts, guiding agents through a controlled forgetting process. Moreover, to evaluate the robustness of the proposed framework, we introduce an unlearning inference adversary capable of crafting prompts, querying agents, and observing their behaviors in an attempt to infer the forgotten knowledge. Experimental results show that our approach effectively enables agents to forget targeted knowledge while preserving performance on untargeted tasks, and prevents the adversary from inferring the forgotten knowledge.

cs.MA↗

Poisoning the Pixels: Revisiting Backdoor Attacks on Semantic Segmentation

Semantic segmentation models are widely deployed in safety-critical applications such as autonomous driving, yet their vulnerability to backdoor attacks remains largely underexplored. Prior segmentation backdoor studies transfer threat settings from existing image classification tasks, focusing primarily on object-to-background mis-segmentation. In this work, we revisit the threats by systematically examining backdoor attacks tailored to semantic segmentation. We identify four coarse-grained attack vectors (Object-to-Object, Object-to-Background, Background-to-Object, and Background-to-Background attacks), as well as two fine-grained vectors (Instance-Level and Conditional attacks). To formalize these attacks, we introduce BADSEG, a unified framework that optimizes trigger designs and applies label manipulation strategies to maximize attack performance while preserving victim model utility. Extensive experiments across diverse segmentation architectures on benchmark datasets demonstrate that BADSEG achieves high attack effectiveness with minimal impact on clean samples. We further evaluate six representative defenses and find that they fail to reliably mitigate our attacks, revealing critical gaps in current defenses. Finally, we demonstrate that these vulnerabilities persist in recent emerging architectures, including transformer-based networks and the Segment Anything Model (SAM), thereby compromising their security. Our work reveals previously overlooked security vulnerabilities in semantic segmentation, and motivates the development of defenses tailored to segmentation-specific threat models.

cs.CR↗

Osmosis Distillation: Model Hijacking with the Fewest Samples

Transfer learning is devised to leverage knowledge from pre-trained models to solve new tasks with limited data and computational resources. Meanwhile, dataset distillation has emerged to synthesize a compact dataset that preserves critical information from the original large dataset. Therefore, a combination of transfer learning and dataset distillation offers promising performance in evaluations. However, a non-negligible security threat remains undiscovered in transfer learning using synthetic datasets generated by dataset distillation methods, where an adversary can perform a model hijacking attack with only a few poisoned samples in the synthetic dataset. To reveal this threat, we propose Osmosis Distillation (OD) attack, a novel model hijacking strategy that targets deep learning models using the fewest samples. Comprehensive evaluations on various datasets demonstrate that the OD attack attains high attack success rates in hidden tasks while preserving high model utility in original tasks. Furthermore, the distilled osmosis set enables model hijacking across diverse model architectures, allowing model hijacking in transfer learning with considerable attack performance and model utility. We argue that awareness of using third-party synthetic datasets in transfer learning must be raised.

cs.CR↗

Chain-of-Lure: A Universal Jailbreak Attack Framework using Unconstrained Synthetic Narratives

In the era of rapid generative AI development, interactions with large language models (LLMs) pose increasing risks of misuse. Prior research has primarily focused on attacks using template-based prompts and optimization-oriented methods, while overlooking the fact that LLMs possess strong unconstrained deceptive capabilities to attack other LLMs. This paper introduces a novel jailbreaking method inspired by the Chain-of-Thought mechanism. The attacker employs mission transfer to conceal harmful user intent within dialogue and generates a progressive chain of lure questions without relying on predefined templates, enabling successful jailbreaks. To further improve the attack's strength, we incorporate a helper LLM model that performs randomized narrative optimization over multi-turn interactions, enhancing the attack performance while preserving alignment with the original intent. We also propose a toxicity-based framework using third-party LLMs to evaluate harmful content and its alignment with malicious intent. Extensive experiments demonstrate that our method consistently achieves high attack success rates and elevated toxicity scores across diverse types of LLMs under black-box API settings. These findings reveal the intrinsic potential of LLMs to perform unrestricted attacks in the absence of robust alignment constraints. Our approach offers data-driven insights to inform the design of future alignment mechanisms. Finally, we propose two concrete defense strategies to support the development of safer generative models. Our code is available at https://github.com/ChangWenhan/chain-of-lure-official

cs.CR↗