Xiangtao Meng

Yingkai Dong, Xiangtao Meng, Ning Yu et al.

Text-to-image (T2I) generative models have revolutionized content creation by transforming textual descriptions into high-quality images. However, these models are vulnerable to jailbreaking attacks, where carefully crafted prompts bypass safety mechanisms to produce unsafe content. While researchers have developed various jailbreak attacks to expose this risk, these methods face significant limitations, including impractical access requirements, easily detectable unnatural prompts, restricted search spaces, and high query demands on the target system. In this paper, we propose JailFuzzer, a novel fuzzing framework driven by large language model (LLM) agents, designed to efficiently generate natural and semantically meaningful jailbreak prompts in a black-box setting. Specifically, JailFuzzer employs fuzz-testing principles with three components: a seed pool for initial and jailbreak prompts, a guided mutation engine for generating meaningful variations, and an oracle function to evaluate jailbreak success. Furthermore, we construct the guided mutation engine and oracle function by LLM-based agents, which further ensures efficiency and adaptability in black-box settings. Extensive experiments demonstrate that JailFuzzer has significant advantages in jailbreaking T2I models. It generates natural and semantically coherent prompts, reducing the likelihood of detection by traditional defenses. Additionally, it achieves a high success rate in jailbreak attacks with minimal query overhead, outperforming existing methods across all key metrics. This study underscores the need for stronger safety mechanisms in generative models and provides a foundation for future research on defending against sophisticated jailbreaking attacks. JailFuzzer is open-source and available at this repository: https://github.com/YingkaiD/JailFuzzer.

Wenyu Chen, Xiangtao Meng, Chuanchao Zang et al.

Large Language Models(LLMs) are widely deployed, yet are vulnerable to jailbreak prompts that elicit policy-violating outputs. Although prior studies have uncovered these risks, they typically treat all tokens as equally important during prompt mutation, overlooking the varying contributions of individual tokens to triggering model refusals. Consequently, these attacks introduce substantial redundant searching under query-constrained scenarios, reducing attack efficiency and hindering comprehensive vulnerability assessment. In this work, we conduct a token-level analysis of refusal behavior and observe that token contributions are highly skewed rather than uniform. Moreover, we find strong cross-model consistency in refusal tendencies, enabling the use of a surrogate model to estimate token-level contributions to the target model's refusals. Motivated by these findings, we propose TriageFuzz, a token-aware jailbreak fuzzing framework that adapts the fuzz testing approach with a series of customized designs. TriageFuzz leverages a surrogate model to estimate the contribution of individual tokens to refusal behaviors, enabling the identification of sensitive regions within the prompt. Furthermore, it incorporates a refusal-guided evolutionary strategy that adaptively weights candidate prompts with a lightweight scorer to steer the evolution toward bypassing safety constraints. Extensive experiments on six open-source LLMs and three commercial APIs demonstrate that TriageFuzz achieves comparable attack success rates (ASR) with significantly reduced query costs. Notably, it attains a 90% ASR with over 70% fewer queries compared to baselines. Even under an extremely restrictive budget of 25 queries, TriageFuzz outperforms existing methods, improving ASR by 20-40%.

Xiangtao Meng, Wenyu Chen, Chuanchao Zang et al.

Large Language Models (LLMs) deployed in high-stakes applications must simultaneously manage multiple risks, yet existing defenses are almost exclusively evaluated in isolation under a one-shot deployment assumption. In practice, providers patch models incrementally throughout their lifecycle-responding to newly exposed vulnerabilities or targeted data-removal requests without retraining from scratch. This raises a fundamental but underexplored question: does a later defense preserve the protections established by an earlier one? We present the first systematic study of cross-defense interactions under sequential deployment. Evaluating 144 ordered sequences across three risk dimensions and three model families, we find that 38.9% exhibit measurable risk exacerbation on the originally defended dimension. These interactions are highly asymmetric and order-dependent. To explain these phenomena, we conduct a mechanistic analysis on representative deployment sequences. Using layer-wise representational divergence and activation patching, we localize each defense to a compact set of critical layers. In conflicting sequences, the overlapping critical layers exhibit strongly anti-aligned parameter updates, whereas benign orderings maintain near-orthogonal updates. PCA trajectory analysis reveals that defense collapse stems from activation pattern reversals in these shared layers. We further introduce a layer-wise conflict score that quantifies the geometric tension between defense-induced activation subspaces, offering mechanistic insight into the observed reversals. Guided by this diagnosis, we propose conflict-guided layer freezing, a lightweight mitigation that selectively freezes high-conflict layers during sequential deployment, preserving prior protections without degrading secondary defense performance.

Xinyu Gao, Xiangtao Meng, Yingkai Dong et al.

While Retrieval-Augmented Generation (RAG) effectively reduces hallucinations by integrating external knowledge bases, it introduces vulnerabilities to membership inference attacks (MIAs), particularly in systems handling sensitive data. Existing MIAs targeting RAG's external databases often rely on model responses but ignore the interference of non-member-retrieved documents on RAG outputs, limiting their effectiveness. To address this, we propose DCMI, a differential calibration MIA that mitigates the negative impact of non-member-retrieved documents. Specifically, DCMI leverages the sensitivity gap between member and non-member retrieved documents under query perturbation. It generates perturbed queries for calibration to isolate the contribution of member-retrieved documents while minimizing the interference from non-member-retrieved documents. Experiments under progressively relaxed assumptions show that DCMI consistently outperforms baselines--for example, achieving 97.42% AUC and 94.35% Accuracy against the RAG system with Flan-T5, exceeding the MBA baseline by over 40%. Furthermore, on real-world RAG platforms such as Dify and MaxKB, DCMI maintains a 10%-20% advantage over the baseline. These results highlight significant privacy risks in RAG systems and emphasize the need for stronger protection mechanisms. We appeal to the community's consideration of deeper investigations, like ours, against the data leakage risks in rapidly evolving RAG systems. Our code is available at https://github.com/Xinyu140203/RAG_MIA.

Xiangtao Meng, Li Wang, Shanqing Guo et al.

While DeepFake applications are becoming popular in recent years, their abuses pose a serious privacy threat. Unfortunately, most related detection algorithms to mitigate the abuse issues are inherently vulnerable to adversarial attacks because they are built atop DNN-based classification models, and the literature has demonstrated that they could be bypassed by introducing pixel-level perturbations. Though corresponding mitigation has been proposed, we have identified a new attribute-variation-based adversarial attack (AVA) that perturbs the latent space via a combination of Gaussian prior and semantic discriminator to bypass such mitigation. It perturbs the semantics in the attribute space of DeepFake images, which are inconspicuous to human beings (e.g., mouth open) but can result in substantial differences in DeepFake detection. We evaluate our proposed AVA attack on nine state-of-the-art DeepFake detection algorithms and applications. The empirical results demonstrate that AVA attack defeats the state-of-the-art black box attacks against DeepFake detectors and achieves more than a 95% success rate on two commercial DeepFake detectors. Moreover, our human study indicates that AVA-generated DeepFake images are often imperceptible to humans, which presents huge security and privacy concerns.

Xiangtao Meng, Yingkai Dong, Ning Yu et al.

Despite the advancements in Text-to-Image (T2I) generation models, their potential for misuse or even abuse raises serious safety concerns. Model developers have made tremendous efforts to introduce safety mechanisms that can address these concerns in T2I models. However, the existing safety mechanisms, whether external or internal, either remain susceptible to evasion under distribution shifts or require extensive model-specific adjustments. To address these limitations, we introduce Safe-Control, an innovative plug-and-play safety patch designed to mitigate unsafe content generation in T2I models. Using data-driven strategies and safety-aware conditions, Safe-Control injects safety control signals into the locked T2I model, acting as an update in a patch-like manner. Model developers can also construct various safety patches to meet the evolving safety requirements, which can be flexibly merged into a single, unified patch. Its plug-and-play design further ensures adaptability, making it compatible with other T2I models of similar denoising architecture. We conduct extensive evaluations on six diverse and public T2I models. Empirical results highlight that Safe-Control is effective in reducing unsafe content generation across six diverse T2I models with similar generative architectures, yet it successfully maintains the quality and text alignment of benign images. Compared to seven state-of-the-art safety mechanisms, including both external and internal defenses, Safe-Control significantly outperforms all baselines in reducing unsafe content generation. For example, it reduces the probability of unsafe content generation to 7%, compared to approximately 20% for most baseline methods, under both unsafe prompts and the latest adversarial attacks.