arXiv subjects

Xinfeng Li

Publications and source records attributed to Xinfeng Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

PromptGuard: Soft Prompt-Guided Unsafe Content Moderation for Text-to-Image Models

Recent text-to-image (T2I) models have exhibited remarkable performance in generating high-quality images from text descriptions. However, these models are vulnerable to misuse, particularly generating not-safe-for-work (NSFW) content, such as sexually explicit, violent, political, and disturbing images, raising serious ethical concerns. In this work, we present PromptGuard, a novel content moderation technique that draws inspiration from the system prompt mechanism in large language models (LLMs) for safety alignment. Unlike LLMs, T2I models lack a direct interface for enforcing behavioral guidelines. Our key idea is to optimize a safety soft prompt that functions as an implicit system prompt within the T2I model's textual embedding space. This universal soft prompt (P*) directly moderates NSFW inputs, enabling safe yet realistic image generation without affecting inference efficiency or requiring proxy models. We further enhance its reliability and helpfulness through a divide-and-conquer strategy that optimizes category-specific soft prompts and combines them into unified safety guidance. Extensive experiments across five datasets demonstrate that PromptGuard effectively mitigates NSFW content generation while preserving high-quality benign outputs. PromptGuard is 3.8 times faster than prior content moderation methods while outperforming eight state-of-the-art defenses. Evaluations using both a multi-head safety classifier and a VLM-based guardrail further confirm its robustness, with average unsafe ratios of 5.84% and 6.18%, respectively. Our code and dataset are available at https://t2i-promptguard.github.io/.

cs.CV↗

arXiv2026

BiasIG: Benchmarking Multi-dimensional Social Biases in Text-to-Image Models

Text-to-Image (T2I) generative models have revolutionized content creation, yet they inherently risk amplifying societal biases. While sociological research provides systematic classifications of bias, existing T2I benchmarks largely conflate these nuances or focus narrowly on occupational stereotypes, leaving the multi-dimensional nature of generative bias inadequately measured. In this paper, we introduce BiasIG, a unified benchmark that quantifies social biases across a curated dataset of 47,040 prompts. Grounded in sociological and machine ethics frameworks, BiasIG disentangles biases across 4 dimensions to enable fine-grained diagnosis. To facilitate scalable and reliable evaluation, we propose a fully automated pipeline powered by a fine-tuned multi-modal large language model, achieving high alignment accuracy comparable to human experts. Extensive experiments on 8 T2I models and 3 debiasing methods not only validate BiasIG as a robust diagnostic tool, but also reveal critical insights: interventions on protected attributes often trigger unintended confounding effects on unrelated demographics, and debiasing methods exhibit a persistent tendency toward discrimination rather than mere ignorance. Our work advocates for a precise, taxonomy-driven approach to fairness in AIGC, providing a theoretical framework for using BiasIG's metrics as feedback signals in future closed-loop mitigation. The benchmark is openly available at https://github.com/Astarojth/BiasIG.

cs.CY↗

arXiv2026

AudioTrust: Benchmarking the Multifaceted Trustworthiness of Audio Large Language Models

The rapid development and widespread adoption of Audio Large Language Models (ALLMs) demand rigorous evaluation of their trustworthiness. However, existing evaluation frameworks are primarily designed for text and fail to capture vulnerabilities introduced by the acoustic properties of audio. We find that significant trustworthiness risks in ALLMs arise from non-semantic acoustic cues, such as timbre, accent, and background noise, which can be exploited to manipulate model behavior. To address this gap, we propose AudioTrust, the first large-scale and systematic framework for evaluating ALLM trustworthiness under audio-specific risks. AudioTrust covers six key dimensions: fairness, hallucination, safety, privacy, robustness, and authenticition. It includes 26 sub-tasks and a curated dataset of more than 4,420 audio samples collected from real-world scenarios, including daily conversations, emergency calls, and voice assistant interactions, and is specifically designed to probe trustworthiness across multiple dimensions. Our comprehensive evaluation spans 18 experimental settings and uses human-validated automated pipelines to enable objective and scalable assessment of model outputs. Experimental results on 14 state-of-the-art open-source and closed-source ALLMs reveal important limitations and failure boundaries under diverse high-risk audio scenarios, providing critical insights for the secure and trustworthy deployment of future audio models. Our platform and benchmark are publicly available at https://github.com/JusperLee/AudioTrust.

cs.SD↗

arXiv2026

You Told Me to Do It: Measuring Instructional Text-induced Private Data Leakage in LLM Agents

High-privilege LLM agents that autonomously process external documentation are increasingly trusted to automate tasks by reading and executing project instructions, yet they are granted terminal access, filesystem control, and outbound network connectivity with minimal security oversight. We identify and systematically measure a fundamental vulnerability in this trust model, which we term the \emph{Trusted Executor Dilemma}: agents execute documentation-embedded instructions, including adversarial ones, at high rates because they cannot distinguish malicious directives from legitimate setup guidance. This vulnerability is a structural consequence of the instruction-following design paradigm, not an implementation bug. To structure our measurement, we formalize a three-dimensional taxonomy covering linguistic disguise, structural obfuscation, and semantic abstraction, and construct \textbf{ReadSecBench}, a benchmark of 500 real-world README files enabling reproducible evaluation. Experiments on the commercially deployed computer-use agent show end-to-end exfiltration success rates up to 85\%, consistent across five programming languages and three injection positions. Cross-model evaluation on four LLM families in a simulation environment confirms that semantic compliance with injected instructions is consistent across model families. A 15-participant user study yields a 0\% detection rate across all participants, and evaluation of 12 rule-based and 6 LLM-based defenses shows neither category achieves reliable detection without unacceptable false-positive rates. Together, these results quantify a persistent \emph{Semantic-Safety Gap} between agents' functional compliance and their security awareness, establishing that documentation-embedded instruction injection is a persistent and currently unmitigated threat to high-privilege LLM agent deployments.

cs.CR↗

arXiv2026

When Memory Becomes a Vulnerability: Towards Multi-turn Jailbreak Attacks against Text-to-Image Generation Systems

Modern text-to-image (T2I) generation systems (e.g., DALL$\cdot$E 3) exploit the memory mechanism, which captures key information in multi-turn interactions for faithful generation. Despite its practicality, the security analyses of this mechanism have fallen far behind. In this paper, we reveal that it can exacerbate the risk of jailbreak attacks. Previous attacks fuse the unsafe target prompt into one ultimate adversarial prompt, which can be easily detected or lead to the generation of non-unsafe images due to under- or over-detoxification. In contrast, we propose embedding the malice at the inception of the chat session in memory, addressing the above limitations. Specifically, we propose Inception, the first multi-turn jailbreak attack against real-world text-to-image generation systems that explicitly exploits their memory mechanisms. Inception is composed of two key modules: segmentation and recursion. We introduce Segmentation, a semantic-preserving method that generates multi-round prompts. By leveraging NLP analysis techniques, we design policies to decompose a prompt, together with its malicious intent, according to sentence structure, thereby evading safety filters. Recursion further addresses the challenge posed by unsafe sub-prompts that cannot be separated through simple segmentation. It firstly expands the sub-prompt, then invokes segmentation recursively. To facilitate multi-turn adversarial prompts crafting, we build VisionFlow, an emulation T2I system that integrates two-stage safety filters and industrial-grade memory mechanisms. The experiment results show that Inception successfully allures unsafe image generation, surpassing the SOTA by a 20.0\% margin in attack success rate. We also conduct experiments on the real-world commercial T2I generation platforms, further validating the threats of Inception in practice.

cs.CV↗

arXiv2026

Mitigating Over-Refusal in Aligned Large Language Models via Inference-Time Activation Energy

Safety alignment of large language models currently faces a central challenge: existing alignment techniques often prioritize mitigating responses to harmful prompts at the expense of overcautious behavior, leading models to incorrectly refuse benign requests. A key goal of safe alignment is therefore to improve safety while simultaneously minimizing false refusals. In this work, we introduce Energy Landscape Steering (ELS), a novel, fine-tuning free framework designed to resolve this challenge through dynamic, inference-time intervention. We train a lightweight external Energy-Based Model (EBM) to assign high energy to undesirable states (false refusal or jailbreak) and low energy to desirable states (helpful response or safe reject). During inference, the EBM maps the LLM's internal activations to an energy landscape, and we use the gradient of the energy function to steer the hidden states toward low-energy regions in real time. This dynamically guides the model toward desirable behavior without modifying its parameters. By decoupling behavioral control from the model's core knowledge, ELS provides a flexible and computationally efficient solution. Extensive experiments across diverse models demonstrate its effectiveness, raising compliance on the ORB-H benchmark from 57.3 percent to 82.6 percent while maintaining baseline safety performance. Our work establishes a promising paradigm for building LLMs that simultaneously achieve high safety and low false refusal rates.

cs.LG↗

arXiv2026

AutoDebias: Automated Framework for Debiasing Text-to-Image Models

Text-to-Image (T2I) models generate high-quality images but are vulnerable to malicious backdoor attacks that inject harmful biases (e.g., trigger-activated gender or racial stereotypes). Existing debiasing methods, often designed for natural statistical biases, struggle with these deliberately and subtly injected attacks. We propose AutoDebias, a framework that automatically identifies and mitigates these malicious biases in T2I models without prior knowledge of the specific attack types. Specifically, AutoDebias leverages vision-language models to detect trigger-activated visual patterns and constructs neutralization guides by generating counter-prompts. These guides drive a CLIP-guided training process that breaks the harmful associations while preserving the original model's image quality and diversity. Unlike methods designed for natural bias, AutoDebias effectively addresses subtle, injected stereotypes and multiple interacting attacks. We evaluate the framework on a new benchmark covering 17 distinct backdoor scenarios, including challenging cases where multiple backdoors co-exist. AutoDebias detects malicious patterns with 91.6% accuracy and reduces the backdoor success rate from 90% to negligible levels, while preserving the visual fidelity of the original model.

cs.CV↗

arXiv2026

"Are You Sure?": An Empirical Study of Human Perception Vulnerability in LLM-Driven Agentic Systems

Large language model (LLM) agents are rapidly becoming trusted copilots in high-stakes domains like software development and healthcare. However, this deepening trust introduces a novel attack surface: Agent-Mediated Deception (AMD), where compromised agents are weaponized against their human users. While extensive research focuses on agent-centric threats, human susceptibility to deception by a compromised agent remains unexplored. We present the first large-scale empirical study with 303 participants to measure human susceptibility to AMD. This is based on HAT-Lab (Human-Agent Trust Laboratory), a high-fidelity research platform we develop, featuring nine carefully crafted scenarios spanning everyday and professional domains (e.g., healthcare, software development, human resources). Our 10 key findings reveal significant vulnerabilities and provide future defense perspectives. Specifically, only 8.6% of participants perceive AMD attacks, while domain experts show increased susceptibility in certain scenarios. We identify six cognitive failure modes in users and find that their risk awareness often fails to translate to protective behavior. The defense analysis reveals that effective warnings should interrupt workflows with low verification costs. With experiential learning based on HAT-Lab, over 90% of users who perceive risks report increased caution against AMD. This work provides empirical evidence and a platform for human-centric agent security research.

cs.HC↗

arXiv2026

The Landscape of Prompt Injection Threats in LLM Agents: From Taxonomy to Analysis

The evolution of Large Language Models (LLMs) has resulted in a paradigm shift towards autonomous agents, necessitating robust security against Prompt Injection (PI) vulnerabilities where untrusted inputs hijack agent behaviors. This SoK presents a comprehensive overview of the PI landscape, covering attacks, defenses, and their evaluation practices. Through a systematic literature review and quantitative analysis, we establish taxonomies that categorize PI attacks by payload generation strategies (heuristic vs. optimization) and defenses by intervention stages (text, model, and execution levels). Our analysis reveals a key limitation shared by many existing defenses and benchmarks: they largely overlook context-dependent tasks, in which agents are authorized to rely on runtime environmental observations to determine actions. To address this gap, we introduce AgentPI, a new benchmark designed to systematically evaluate agent behavior under context-dependent interaction settings. Using AgentPI, we empirically evaluate representative defenses and show that no single approach can simultaneously achieve high trustworthiness, high utility, and low latency. Moreover, we show that many defenses appear effective under existing benchmarks by suppressing contextual inputs, yet fail to generalize to realistic agent settings where context-dependent reasoning is essential. This SoK distills key takeaways and open research problems, offering structured guidance for future research and practical deployment of secure LLM agents.

cs.CR↗

arXiv2026

AgentAuditor: Human-Level Safety and Security Evaluation for LLM Agents

Despite the rapid advancement of LLM-based agents, the reliable evaluation of their safety and security remains a significant challenge. Existing rule-based or LLM-based evaluators often miss dangers in agents' step-by-step actions, overlook subtle meanings, fail to see how small issues compound, and get confused by unclear safety or security rules. To overcome this evaluation crisis, we introduce AgentAuditor, a universal, training-free, memory-augmented reasoning framework that empowers LLM evaluators to emulate human expert evaluators. AgentAuditor constructs an experiential memory by having an LLM adaptively extract structured semantic features (e.g., scenario, risk, behavior) and generate associated chain-of-thought reasoning traces for past interactions. A multi-stage, context-aware retrieval-augmented generation process then dynamically retrieves the most relevant reasoning experiences to guide the LLM evaluator's assessment of new cases. Moreover, we developed ASSEBench, the first benchmark designed to check how well LLM-based evaluators can spot both safety risks and security threats. ASSEBench comprises 2293 meticulously annotated interaction records, covering 15 risk types across 29 application scenarios. A key feature of ASSEBench is its nuanced approach to ambiguous risk situations, employing "Strict" and "Lenient" judgment standards. Experiments demonstrate that AgentAuditor not only consistently improves the evaluation performance of LLMs across all benchmarks but also sets a new state-of-the-art in LLM-as-a-judge for agent safety and security, achieving human-level accuracy. Our work is openly accessible at https://github.com/Astarojth/AgentAuditor.

cs.AI↗

arXiv2025

ThreatPilot: Attack-Driven Threat Intelligence Extraction

Efficient defense against dynamically evolving advanced persistent threats (APT) requires the structured threat intelligence feeds, such as techniques used. However, existing threat-intelligence extraction techniques predominantly focuses on individual pieces of intelligence-such as isolated techniques or atomic indicators-resulting in fragmented and incomplete representations of real-world attacks. This granularity inherently limits on both the depth and the contextual richness of the extracted intelligence, making it difficult for downstream security systems to reason about multi-step behaviors or to generate actionable detections. To address this gap, we propose to extract the layered Attack-driven Threat Intelligence (ATIs), a comprehensive representation that captures the full spectrum of adversarial behavior. We propose ThreatPilot, which can accurately identify the AITs including complete tactics, techniques, multi-step procedures, and their procedure variants, and integrate the threat intelligence to software security application scenarios: the detection rules (i.e., Sigma) and attack command can be generated automatically to a more accuracy level. Experimental results on 1,769 newly crawly reports and 16 manually calibrated reports show ThreatPilot's effectiveness in identifying accuracy techniques, outperforming state-of-the-art approaches of AttacKG by 1.34X in F1 score. Further studies upon 64,185 application logs via Honeypot show that our Sigma rule generator significantly outperforms several existing rules-set in detecting the real-world malicious events. Industry partners confirm that our Sigma rule generator can significantly help save time and costs of the rule generation process. In addition, our generated commands achieve an execution rate of 99.3%, compared to 50.3% without the extracted intelligence.

cs.CR↗

arXiv2025

MCI: Multi-Channel Imager on the Chinese Space Station Survey Telescope

The Multi-Channel Imager (MCI) is a powerful near-ultraviolet (NUV) and visible imager onboard the Chinese Space Station Survey Telescope (CSST). The MCI provides three imaging channels, which are the NUV channel, the Blue channel and the Red channel, with the wavelength range of 255-430 nm, 430-700 nm, and 700-1000 nm, respectively. MCI's three channels can target the same field simultaneously, which is unique compared to other imagers onboard the Hubble Space Telescope (HST) or the James Webb Space Telescope (JWST). Each channel employs a CCD focal plane of 9216 x 9232 pixels and $\sim$7\arcmin.5 x 7\arcmin.5 field of view (FOV), which are about $\gtrsim 4$ times greater than the FOVs of HST imagers. The MCI's three channels feature unprecedented sensitivities and field of views complement the NUV and visible capabilities of the CSST for high-precision photometry and weak-signal detection, which would help build a new standard-star system and the deepest UV-Optical exposures for CSST. Rich filter sets of MCI would help explore other sciences such as local emission line mapping, high-z Ly$α$ emitters searching, etc. Here we present key design features, results of current ground tests, and suggested observing strategies of the MCI.

astro-ph.IM↗

arXiv2025

EmoRAG: Evaluating RAG Robustness to Symbolic Perturbations

Retrieval-Augmented Generation (RAG) systems are increasingly central to robust AI, enhancing large language model (LLM) faithfulness by incorporating external knowledge. However, our study unveils a critical, overlooked vulnerability: their profound susceptibility to subtle symbolic perturbations, particularly through near-imperceptible emoticon tokens such as "(@_@)" that can catastrophically mislead retrieval, termed EmoRAG. We demonstrate that injecting a single emoticon into a query makes it nearly 100% likely to retrieve semantically unrelated texts that contain a matching emoticon. Our extensive experiment across general question-answering and code domains, using a range of state-of-the-art retrievers and generators, reveals three key findings: (I) Single-Emoticon Disaster: Minimal emoticon injections cause maximal disruptions, with a single emoticon almost 100% dominating RAG output. (II) Positional Sensitivity: Placing an emoticon at the beginning of a query can cause severe perturbation, with F1-Scores exceeding 0.92 across all datasets. (III) Parameter-Scale Vulnerability: Counterintuitively, models with larger parameters exhibit greater vulnerability to the interference. We provide an in-depth analysis to uncover the underlying mechanisms of these phenomena. Furthermore, we raise a critical concern regarding the robustness assumption of current RAG systems, envisioning a threat scenario where an adversary exploits this vulnerability to manipulate the RAG system. We evaluate standard defenses and find them insufficient against EmoRAG. To address this, we propose targeted defenses, analyzing their strengths and limitations in mitigating emoticon-based perturbations. Finally, we outline future directions for building robust RAG systems.

cs.CR↗

arXiv2025

EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models

Many machine learning models are fine-tuned from large language models (LLMs) to achieve high performance in specialized domains like code generation, biomedical analysis, and mathematical problem solving. However, this fine-tuning process often introduces a critical vulnerability: the systematic degradation of safety alignment, undermining ethical guidelines and increasing the risk of harmful outputs. Addressing this challenge, we introduce EnchTable, a novel framework designed to transfer and maintain safety alignment in downstream LLMs without requiring extensive retraining. EnchTable leverages a Neural Tangent Kernel (NTK)-based safety vector distillation method to decouple safety constraints from task-specific reasoning, ensuring compatibility across diverse model architectures and sizes. Additionally, our interference-aware merging technique effectively balances safety and utility, minimizing performance compromises across various task domains. We implemented a fully functional prototype of EnchTable on three different task domains and three distinct LLM architectures, and evaluated its performance through extensive experiments on eleven diverse datasets, assessing both utility and model safety. Our evaluations include LLMs from different vendors, demonstrating EnchTable's generalization capability. Furthermore, EnchTable exhibits robust resistance to static and dynamic jailbreaking attacks, outperforming vendor-released safety models in mitigating adversarial prompts. Comparative analyses with six parameter modification methods and two inference-time alignment baselines reveal that EnchTable achieves a significantly lower unsafe rate, higher utility score, and universal applicability across different task domains. Additionally, we validate EnchTable can be seamlessly integrated into various deployment pipelines without significant overhead.

cs.CL↗

arXiv2025

A Vision for Access Control in LLM-based Agent Systems

The autonomy and contextual complexity of LLM-based agents render traditional access control (AC) mechanisms insufficient. Static, rule-based systems designed for predictable environments are fundamentally ill-equipped to manage the dynamic information flows inherent in agentic interactions. This position paper argues for a paradigm shift from binary access control to a more sophisticated model of information governance, positing that the core challenge is not merely about permission, but about governing the flow of information. We introduce Agent Access Control (AAC), a novel framework that reframes AC as a dynamic, context-aware process of information flow governance. AAC operates on two core modules: (1) multi-dimensional contextual evaluation, which assesses not just identity but also relationships, scenarios, and norms; and (2) adaptive response formulation, which moves beyond simple allow/deny decisions to shape information through redaction, summarization, and paraphrasing. This vision, powered by a dedicated AC reasoning engine, aims to bridge the gap between human-like nuanced judgment and scalable Al safety, proposing a new conceptual lens for future research in trustworthy agent design.

cs.MA↗

arXiv2025

Longwave-transparent low-emissivity material

Low emissivity (low-e) materials are crucial for conserving thermal energy in buildings, cold chain logistics and transportation by minimizing unwanted radiative heat loss or gain. However, their metallic nature intrinsically causes severe longwave attenuation, hindering their broad applications. Here, we introduce, for the first time, an all-dielectric longwave-transparent low-emissivity material (LLM) with ultra-broadband, high transmittance spanning 9 orders of magnitude, from terahertz to kilohertz frequencies. This meter-scale LLM not only achieves energy savings of up to 41.1% over commercial white paint and 10.2% over traditional low-e materials, but also unlocks various fundamentally new capabilities including high-speed wireless communication in energy-efficient buildings, wireless energy transfer with radiative thermal insulation, as well as non-invasive terahertz security screening and radio frequency identification in cold chain logistics. Our approach represents a new photonic solution towards carbon neutrality and smart city development, paving the way for a more sustainable and interconnected future.

physics.optics↗

arXiv2025

Patronus: Safeguarding Text-to-Image Models against White-Box Adversaries

Text-to-image (T2I) models, though exhibiting remarkable creativity in image generation, can be exploited to produce unsafe images. Existing safety measures, e.g., content moderation or model alignment, fail in the presence of white-box adversaries who know and can adjust model parameters, e.g., by fine-tuning. This paper presents a novel defensive framework, named Patronus, which equips T2I models with holistic protection to defend against white-box adversaries. Specifically, we design an internal moderator that decodes unsafe input features into zero vectors while ensuring the decoding performance of benign input features. Furthermore, we strengthen the model alignment with a carefully designed non-fine-tunable learning mechanism, ensuring the T2I model will not be compromised by malicious fine-tuning. We conduct extensive experiments to validate the intactness of the performance on safe content generation and the effectiveness of rejecting unsafe content generation. Results also confirm the resilience of Patronus against various fine-tuning attacks by white-box adversaries.

cs.CR↗

arXiv2025

SeCon-RAG: A Two-Stage Semantic Filtering and Conflict-Free Framework for Trustworthy RAG

Retrieval-augmented generation (RAG) systems enhance large language models (LLMs) with external knowledge but are vulnerable to corpus poisoning and contamination attacks, which can compromise output integrity. Existing defenses often apply aggressive filtering, leading to unnecessary loss of valuable information and reduced reliability in generation. To address this problem, we propose a two-stage semantic filtering and conflict-free framework for trustworthy RAG. In the first stage, we perform a joint filter with semantic and cluster-based filtering which is guided by the Entity-intent-relation extractor (EIRE). EIRE extracts entities, latent objectives, and entity relations from both the user query and filtered documents, scores their semantic relevance, and selectively adds valuable documents into the clean retrieval database. In the second stage, we proposed an EIRE-guided conflict-aware filtering module, which analyzes semantic consistency between the query, candidate answers, and retrieved knowledge before final answer generation, filtering out internal and external contradictions that could mislead the model. Through this two-stage process, SeCon-RAG effectively preserves useful knowledge while mitigating conflict contamination, achieving significant improvements in both generation robustness and output trustworthiness. Extensive experiments across various LLMs and datasets demonstrate that the proposed SeCon-RAG markedly outperforms state-of-the-art defense methods.

cs.CL↗