Advanced deepfake technologies are blurring the lines between real and fake, presenting both revolutionary opportunities and alarming threats. While it unlocks novel applications in fields like entertainment and education, its malicious use has sparked urgent ethical and societal concerns ranging from identity theft to the dissemination of misinformation. To tackle these challenges, feature analysis using frequency features has emergedas a promising direction for deepfake detection. However, oneaspect that has been overlooked so far is that existing methodstend to concentrate on one or a few specific frequency domains,which risks overfitting to particular artifacts and significantlyundermines their robustness when facing diverse forgery patterns. Another underexplored aspect we observe is that different features often attend to the same forged region, resulting in redundant feature representations and limiting the diversity of the extracted clues. This may undermine the ability of a model to capture complementary information across different facets, thereby compromising its generalization capability to diverse manipulations. In this paper, we seek to tackle these challenges from two aspects: (1) we propose a triple-branch network that jointly captures spatial and frequency features by learning from both original image and image reconstructed by different frequency channels, and (2) we mathematically derive feature decoupling and fusion losses grounded in the mutual information theory, which enhances the model to focus on task-relevant features across the original image and the image reconstructed by different frequency channels. Extensive experiments on six large-scale benchmark datasets demonstrate that our method consistently achieves state-of-the-art performance. Our code is released at https://github.com/injooker/Unveiling Deepfake.

Modern Large audio-language models (LALMs) power intelligent voice interactions by tightly integrating audio and text. This integration, however, expands the attack surface beyond text and introduces vulnerabilities in the continuous, high-dimensional audio channel. While prior work studied audio jailbreaks, the security risks of malicious audio injection and downstream behavior manipulation remain underexamined. In this work, we reveal a previously overlooked threat, auditory prompt injection, under realistic constraints of audio data-only access and strong perceptual stealth. To systematically analyze this threat, we propose \textit{AudioHijack}, a general framework that generates context-agnostic and imperceptible adversarial audio to hijack LALMs. \textit{AudioHijack} employs sampling-based gradient estimation for end-to-end optimization across diverse models, bypassing non-differentiable audio tokenization. Through attention supervision and multi-context training, it steers model attention toward adversarial audio and generalizes to unseen user contexts. We also design a convolutional blending method that modulates perturbations into natural reverberation, making them highly imperceptible to users. Extensive experiments on 13 state-of-the-art LALMs show consistent hijacking across 6 misbehavior categories, achieving average success rates of 79\%-96\% on unseen user contexts with high acoustic fidelity. Real-world studies demonstrate that commercial voice agents from Mistral AI and Microsoft Azure can be induced to execute unauthorized actions on behalf of users. These findings expose critical vulnerabilities in LALMs and highlight the urgent need for dedicated defense.

Ensuring Large Language Model (LLM) safety is crucial, yet the lack of a clear understanding about safety mechanisms hinders the development of precise and reliable methodologies for safety intervention across diverse tasks. To better understand and control LLM safety, we propose the Expected Safety Impact (ESI) framework for quantifying how different parameters affect LLM safety. Based on ESI, we reveal distinct safety-critical patterns across different LLM architectures: In dense LLMs, many safety-critical parameters are located in value matrices (V) and MLPs in middle layers, whereas in Mixture-of-Experts (MoE) models, they shift to the late-layer MLPs. Leveraging ESI, we further introduce two targeted intervention paradigms for safety enhancement and preservation, i.e., Safety Enhancement Tuning (SET) and Safety Preserving Adaptation (SPA). SET can align unsafe LLMs by updating only a few safety-critical parameters, effectively enhancing safety while preserving original performance. SPA safeguards well-aligned LLMs during capability-oriented intervention (e.g., instruction tuning) by preventing disruption of safety-critical weights, allowing the LLM to acquire new abilities and maintain safety capabilities. Extensive evaluations on different LLMs demonstrate that SET can reduce the attack success rates of unaligned LLMs by over 50% with only a 100-iteration update on 1% of model weights. SPA can limit the safety degradation of aligned LLMs within 1% after a 1,000-iteration instruction fine-tuning on different tasks. Our code is available at: https://github.com/ZJU-LLM-Safety/SafeWeights-ACL.

Advances in AIGC technologies have enabled the synthesis of highly realistic audio deepfakes capable of deceiving human auditory perception. Although numerous audio deepfake detection (ADD) methods have been developed, most rely on local temporal/spectral features or pairwise relations, overlooking high-order interactions (HOIs). HOIs capture discriminative patterns that emerge from multiple feature components beyond their individual contributions. We propose HyperPotter, a hypergraph-based framework that explicitly models these synergistic HOIs through clustering-based hyperedges with class-aware prototype initialization. Extensive experiments demonstrate that HyperPotter surpasses its baseline by an average relative gain of 22.15% across 11 datasets and outperforms state-of-the-art methods by 13.96% on 4 challenging cross-domain datasets, demonstrating superior generalization to diverse attacks and speakers.

Large language models (LLMs) for Verilog code generation are increasingly adopted in hardware design, yet remain vulnerable to backdoor attacks where adversaries inject malicious triggers during training to induce vulnerable hardware designs. Unlike patchable software vulnerabilities, hardware trojans become irreversible once fabricated, making remediation extremely costly or impossible. Existing active defenses require access to training data, impractical for third-party LLM users, while passive defenses struggle against semantically stealthy triggers that naturally blend into design specifications. In this paper, we hypothesize that under the requirements of both effectiveness and stealthiness, attackers are strongly biased toward embedding triggers in non-functional requirements (e.g., style modifiers, quality descriptors) rather than functional specifications that determine hardware behavior. Exploiting this insight, we propose Semantic Consensus Decoding (SCD), an inference-time passive defense with two key components: (1) functional requirement extraction that identifies essential requirements from user specifications, and (2) consensus decoding that adaptively fuses output distributions based on full user specifications and extracted functional requirements. When these distributions diverge significantly, SCD automatically suppresses suspicious components. Extensive experiments with three representative backdoor attacks demonstrate that SCD reduces average attack success rate from 89% to under 3% with negligible impact on generation quality.

Existing gradient-based jailbreak attacks on Large Language Models (LLMs) typically optimize adversarial suffixes to align the LLM output with predefined target responses. However, restricting the objective as inducing fixed targets inherently constrains the adversarial search space, limiting the overall attack efficacy. Furthermore, existing methods typically require numerous optimization iterations to fulfill the large gap between the fixed target and the original LLM output, resulting in low attack efficiency. To overcome these limitations, we propose the first gradient-based untargeted jailbreak attack (UJA), which relies on an untargeted objective to maximize the unsafety probability of the LLM output, without enforcing any response patterns. For tractable optimization, we further decompose this objective into two differentiable sub-objectives to search the optimal harmful response and the corresponding adversarial prompt, with a theoretical analysis to validate the decomposition. In contrast to existing attacks, UJA's unrestricted objective significantly expands the search space, enabling more flexible and efficient exploration of LLM vulnerabilities. Extensive evaluations show that UJA achieves over 80\% attack success rates against recent safety-aligned LLMs with only 100 optimization iterations, outperforming the state-of-the-art gradient-based attacks by over 30\%.

In recent years, RAG has emerged as a key paradigm for enhancing large language models (LLMs). By integrating externally retrieved information, RAG alleviates issues like outdated knowledge and, crucially, insufficient domain expertise. While effective, RAG introduces new risks of external data extraction attacks (EDEAs), where sensitive or copyrighted data in its knowledge base may be extracted verbatim. These risks are particularly acute when RAG is used to customize specialized LLM applications with private knowledge bases. Despite initial studies exploring these risks, they often lack a formalized framework, robust attack performance, and comprehensive evaluation, leaving critical questions about real-world EDEA feasibility unanswered. In this paper, we present the first comprehensive study to formalize EDEAs against retrieval-augmented LLMs. We first formally define EDEAs and propose a unified framework decomposing their design into three components: extraction instruction, jailbreak operator, and retrieval trigger, under which prior attacks can be considered instances within our framework. Guided by this framework, we develop SECRET: a Scalable and EffeCtive exteRnal data Extraction aTtack. Specifically, SECRET incorporates (1) an adaptive optimization process using LLMs as optimizers to generate specialized jailbreak prompts for EDEAs, and (2) cluster-focused triggering, an adaptive strategy that alternates between global exploration and local exploitation to efficiently generate effective retrieval triggers. Extensive evaluations across 4 models reveal that SECRET significantly outperforms previous attacks, and is highly effective against all 16 tested RAG instances. Notably, SECRET successfully extracts 35% of the data from RAG powered by Claude 3.7 Sonnet for the first time, whereas other attacks yield 0% extraction. Our findings call for attention to this emerging threat.

Existing gradient-based jailbreak attacks typically optimize an adversarial suffix to induce a fixed affirmative response, e.g., ``Sure, here is...''. However, this fixed target usually resides in an extremely low-density region of a safety-aligned LLM's output distribution. Due to the substantial discrepancy between the fixed target and the output distribution, existing attacks require numerous iterations to optimize the adversarial prompt, which might still fail to induce the low-probability target response. To address this limitation, we propose Dynamic Target Attack (DTA), which leverages the target LLM's own responses as adaptive targets. In each optimization round, DTA samples multiple candidates from the output distribution conditioned on the current prompt, and selects the most harmful one as a temporary target for prompt optimization. Extensive experiments demonstrate that, under the white-box setting, DTA achieves over 87% average attack success rate (ASR) within 200 optimization iterations on recent safety-aligned LLMs, exceeding the state-of-the-art baselines by over 15% and reducing wall-clock time by 2-26x. Under the black-box setting, DTA employs a white-box LLM as a surrogate model for gradient-based optimization, achieving an average ASR of 77.5% against black-box models, exceeding prior transfer-based attacks by over 12%.

While the pursuit of higher accuracy in deepfake detection remains a central goal, there is an increasing demand for precise localization of manipulated regions. Despite the remarkable progress made in classification-based detection, accurately localizing forged areas remains a significant challenge. A common strategy is to incorporate forged region annotations during model training alongside manipulated images. However, such approaches often neglect the complementary nature of local detail and global semantic context, resulting in suboptimal localization performance. Moreover, an often-overlooked aspect is the fusion strategy between local and global predictions. Naively combining the outputs from both branches can amplify noise and errors, thereby undermining the effectiveness of the localization. To address these issues, we propose a novel approach that independently predicts manipulated regions using both local and global perspectives. We employ morphological operations to fuse the outputs, effectively suppressing noise while enhancing spatial coherence. Extensive experiments reveal the effectiveness of each module in improving the accuracy and robustness of forgery localization.

The broad capabilities and substantial resources required to train Large Language Models (LLMs) make them valuable intellectual property, yet they remain vulnerable to copyright infringement, such as unauthorized use and model theft. LLM fingerprinting, a non-intrusive technique that compares the distinctive features (i.e., fingerprint) of LLMs to identify whether an LLM is derived from another, offers a promising solution to copyright auditing. However, its reliability remains uncertain due to the prevalence of diverse model modifications and the lack of standardized evaluation. In this SoK, we present the first comprehensive study of the emerging LLM fingerprinting. We introduce a unified framework and taxonomy that structures the field: white-box methods are classified based on their feature source as static, forward-pass, or backward-pass fingerprinting, while black-box methods are distinguished by their query strategy as either untargeted or targeted. Furthermore, we propose LeaFBench, the first systematic benchmark for evaluating LLM fingerprinting under realistic deployment scenarios. Built upon 7 mainstream foundation models and comprising 149 distinct model instances, LeaFBench integrates 13 representative post-development techniques, spanning both parameter-altering methods (e.g., fine-tuning, quantization) and parameter-independent techniques (e.g., system prompts, RAG). Extensive experiments on LeaFBench reveal the strengths and weaknesses of existing methods, thereby outlining future research directions and critical open problems in this emerging field. The code is available at https://github.com/shaoshuo-ss/LeaFBench.

Large Language Models (LLMs) have exhibited remarkable capabilities but remain vulnerable to jailbreaking attacks, which can elicit harmful content from the models by manipulating the input prompts. Existing black-box jailbreaking techniques primarily rely on static prompts crafted with a single, non-adaptive strategy, or employ rigid combinations of several underperforming attack methods, which limits their adaptability and generalization. To address these limitations, we propose MAJIC, a Markovian adaptive jailbreaking framework that attacks black-box LLMs by iteratively combining diverse innovative disguise strategies. MAJIC first establishes a ``Disguise Strategy Pool'' by refining existing strategies and introducing several innovative approaches. To further improve the attack performance and efficiency, MAJIC formulate the sequential selection and fusion of strategies in the pool as a Markov chain. Under this formulation, MAJIC initializes and employs a Markov matrix to guide the strategy composition, where transition probabilities between strategies are dynamically adapted based on attack outcomes, thereby enabling MAJIC to learn and discover effective attack pathways tailored to the target model. Our empirical results demonstrate that MAJIC significantly outperforms existing jailbreak methods on prominent models such as GPT-4o and Gemini-2.0-flash, achieving over 90\% attack success rate with fewer than 15 queries per attempt on average.

The Key-Value (KV) cache, which stores intermediate attention computations (Key and Value pairs) to avoid redundant calculations, is a fundamental mechanism for accelerating Large Language Model (LLM) inference. However, this efficiency optimization introduces significant yet underexplored privacy risks. This paper provides the first comprehensive analysis of these vulnerabilities, demonstrating that an attacker can reconstruct sensitive user inputs directly from the KV-cache. We design and implement three distinct attack vectors: a direct Inversion Attack, a more broadly applicable and potent Collision Attack, and a semantic-based Injection Attack. These methods demonstrate the practicality and severity of KV-cache privacy leakage issues. To mitigate this, we propose KV-Cloak, a novel, lightweight, and efficient defense mechanism. KV-Cloak uses a reversible matrix-based obfuscation scheme, combined with operator fusion, to secure the KV-cache. Our extensive experiments show that KV-Cloak effectively thwarts all proposed attacks, reducing reconstruction quality to random noise. Crucially, it achieves this robust security with virtually no degradation in model accuracy and minimal performance overhead, offering a practical solution for trustworthy LLM deployment.