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.

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.

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.

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.

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.

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 raises the defense success rate from a baseline of 0.32 to over 0.89 and significantly mitigates the cascading spread of minor errors.

Turning Black Box into White Box: Dataset Distillation Leaks

Dataset distillation compresses a large real dataset into a small synthetic one, enabling models trained on the synthetic data to achieve performance comparable to those trained on the real data. Although synthetic datasets are assumed to be privacy-preserving, we show that existing distillation methods can cause severe privacy leakage because synthetic datasets implicitly encode the weight trajectories of the distilled model, they become over-informative and exploitable by adversaries. To expose this risk, we introduce the Information Revelation Attack (IRA) against state-of-the-art distillation techniques. Experiments show that IRA accurately predicts both the distillation algorithm and model architecture, and can successfully infer membership and recover sensitive samples from the real dataset.

Hide&Seek: Remove Image Watermarks with Negligible Cost via Pixel-wise Reconstruction

Watermarking has emerged as a key defense against the misuse of machine-generated images (MGIs). Yet the robustness of these protections remains underexplored. To reveal the limits of SOTA proactive image watermarking defenses, we propose HIDE&SEEK (HS), a suite of versatile and cost-effective attacks that reliably remove embedded watermarks while preserving high visual fidelity.

Guided Collaboration in Heterogeneous LLM-Based Multi-Agent Systems via Entropy-Based Understanding Assessment and Experience Retrieval

With recent breakthroughs in large language models (LLMs) for reasoning, planning, and complex task generation, artificial intelligence systems are transitioning from isolated single-agent architectures to multi-agent systems with collaborative intelligence. However, in heterogeneous multi-agent systems (HMAS), capability differences among agents give rise to consistent cognitive problems, where strong and weak models fail to contribute effectively. We define the collaboration as a strong-weak system. Through comprehensive experiments, we disclose a counterintuitive phenomenon in the strong-weak system: a strong-weak collaboration may under-perform weak-weak combinations, revealing that cognitive mismatching are key bottlenecks limiting heterogeneous cooperation. To overcome these challenges, we propose an Entropy-Based Adaptive Guidance Framework that dynamically aligns the guidance with the cognitive state of each agent. The framework quantifies the understanding of weak agents through multi-dimensional entropy metrics - covering expression, uncertainty, structure, coherence, and relevance - and adaptively adjusts the intensity of the guidance at light, moderate and intensive levels. Furthermore, a Retrieval-Augmented Generation (RAG) mechanism is incorporated to retain successful collaboration experiences, enabling both immediate adaptation and long-term learning. Extensive experiments on three benchmark datasets, GSM8K, MBPP, and CVRP demonstrate that our approach consistently enhances the effectiveness and stability of heterogeneous collaboration. The results highlight that adaptive guidance not only mitigates cognitive imbalance but also establishes a scalable pathway toward more robust, cooperative multi-agent intelligence.

Forgetting Similar Samples: Can Machine Unlearning Do it Better?

Machine unlearning, a process enabling pre-trained models to remove the influence of specific training samples, has attracted significant attention in recent years. Although extensive research has focused on developing efficient machine unlearning strategies, we argue that these methods mainly aim at removing samples rather than removing samples' influence on the model, thus overlooking the fundamental definition of machine unlearning. In this paper, we first conduct a comprehensive study to evaluate the effectiveness of existing unlearning schemes when the training dataset includes many samples similar to those targeted for unlearning. Specifically, we evaluate: Do existing unlearning methods truly adhere to the original definition of machine unlearning and effectively eliminate all influence of target samples when similar samples are present in the training dataset? Our extensive experiments, conducted on four carefully constructed datasets with thorough analysis, reveal a notable gap between the expected and actual performance of most existing unlearning methods for image and language models, even for the retraining-from-scratch baseline. Additionally, we also explore potential solutions to enhance current unlearning approaches.