As large language models (LLMs) are increasingly deployed, ensuring their safe use is paramount. Jailbreaking, adversarial prompts that bypass model alignment to trigger harmful outputs, present significant risks, with existing studies reporting high success rates in evading common LLMs. However, previous evaluations have focused solely on the models, neglecting the full deployment pipeline, which typically incorporates additional safety mechanisms like content moderation filters. To address this gap, we present the first systematic evaluation of jailbreak attacks targeting LLM safety alignment, assessing their success across the full inference pipeline, including both input and output filtering stages. Our findings yield two key insights: first, nearly all evaluated jailbreak techniques can be detected by at least one safety filter, suggesting that prior assessments may have overestimated the practical success of these attacks; second, while safety filters are effective in detection, there remains room to better balance recall and precision to further optimize protection and user experience. We highlight critical gaps and call for further refinement of detection accuracy and usability in LLM safety systems.

Recently, multimodal large language models (MLLMs) have emerged as a unified paradigm for language and image generation. Our work shows that MLLMs pair usability with higher risks, highlighting the need for adaptive safeguards to mitigate real-world harms.Compared with diffusion models, MLLMs possess a much stronger capability for semantic understanding, enabling them to process more complex textual inputs and comprehend richer contextual meanings. However, this enhanced semantic ability may also introduce new and potentially greater safety risks.Taking diffusion models as a reference point, we systematically analyze and compare the safety risks of emerging MLLMs along two dimensions: unsafe content generation and fake image synthesis.Across multiple unsafe generation benchmark datasets, we observe that MLLMs tend to generate more unsafe images than diffusion models. This difference partly arises because diffusion models often fail to interpret abstract prompts, producing corrupted outputs, whereas MLLMs can comprehend these prompts and generate unsafe content.For current advanced fake image detectors, MLLM-generated images are also notably harder to identify. Even when detectors are retrained with MLLMs-specific data, they can still be bypassed by simply providing MLLMs with longer and more descriptive inputs.Our measurements indicate that the emerging safety risks of the cutting-edge generative paradigm, MLLMs, have not been sufficiently recognized, posing new challenges to real-world safety.

Transferable adversarial images raise critical security concerns for computer vision systems in real-world, black-box attack scenarios. Although many transfer attacks have been proposed, existing research lacks a systematic and comprehensive evaluation. In this paper, we systemize transfer attacks into five categories around the general machine learning pipeline and provide the first comprehensive evaluation, with 23 representative attacks against 11 representative defenses, including the recent, transfer-oriented defense and the real-world Google Cloud Vision. In particular, we identify two main problems of existing evaluations: (1) for attack transferability, lack of intra-category analyses with fair hyperparameter settings, and (2) for attack stealthiness, lack of diverse measures. Our evaluation results validate that these problems have indeed caused misleading conclusions and missing points, and addressing them leads to new, consensus-challenging insights, such as (1) an early attack, DI, even outperforms all similar follow-up ones, (2) the state-of-the-art (white-box) defense, DiffPure, is even vulnerable to (black-box) transfer attacks, and (3) even under the same $L_{p}$Lp constraint, different attacks yield dramatically different stealthiness results regarding diverse imperceptibility metrics, finer-grained measures, and a user study. We hope that our analyses will serve as guidance on properly evaluating transferable adversarial images and advance the design of attacks and defenses.

Large language models (LLMs) are powerful at question-answering but prone to hallucinations due to limited domain-specific or up-to-date knowledge. Retrieval augmented generation (RAG) mitigates this by adding an external retriever and knowledge database, yet RAG remains vulnerable to targeted attacks that degrade outputs or manipulate opinions. Prior attacks typically assume adversaries know the service is RAG-enhanced and may even know deployment details, an assumption often invalid for real-world commercial LLMs that expose only black-box APIs.This opacity also risks misleading users about system capabilities. This work aims to bridge this gap by proposing RAG-ID , a framework for  ̲ ID entifying  ̲ RAG properties in LLM services.We classify adversaries into three knowledge levels and design six attack methods. Experiments show these attacks reliably detect RAG — up to 99.97% accuracy with partial or no optional knowledge, and nearly 100% when the LLM and database are known. After detection, RAG-ID can infer finer RAG properties (e.g., deployed LLM and knowledge database). We consider RAG-ID a reconnaissance tool for attackers, a way to facilitate users’ transparent selection of LLM services, and a guide for RAG developers in refining security measures.

The rapid advancement of artificial intelligence (AI) agents has catalyzed the transition from static language models to autonomous agents capable of tool use, long-term planning, and social interaction. Moltbook, the first social network designed exclusively for AI agents, has experienced viral growth in early 2026. To understand the behavior of AI agents in the agent-native community, in this paper, we present a large-scale empirical analysis of Moltbook leveraging a dataset of 44,411 posts and 12,209 sub-communities ("submolts") collected prior to February 1, 2026. Leveraging a topic taxonomy with nine content categories and a five-level toxicity scale, we systematically analyze the topics and risks of agent discussions. Our analysis answers three questions: what topics do agents discuss (RQ1), how risk varies by topic (RQ2), and how topics and toxicity evolve over time (RQ3). We find that Moltbook exhibits explosive growth and rapid diversification, moving beyond early social interaction into viewpoint, incentive-driven, promotional, and political discourse. The attention of agents increasingly concentrates in centralized hubs and around polarizing, platform-native narratives. Toxicity is strongly topic-dependent: incentive- and governance-centric categories contribute a disproportionate share of risky content, including religion-like coordination rhetoric and anti-humanity ideology. Moreover, bursty automation by a small number of agents can produce flooding at sub-minute intervals, distorting discourse and stressing platform stability. Overall, our study underscores the need for topic-sensitive monitoring and platform-level safeguards in agent social networks.

By introducing routers to selectively activate experts in Transformer layers, the mixture-of-experts (MoE) architecture significantly reduces computational costs in large language models (LLMs) while maintaining competitive performance, especially for models with massive parameters. However, prior work has largely focused on utility and efficiency, leaving the safety risks associated with this sparse architecture underexplored. In this work, we show that the safety of MoE LLMs is as sparse as their architecture by discovering unsafe routes: routing configurations that, once activated, convert safe outputs into harmful ones. Specifically, we first introduce the Router Safety importance score (RoSais) to quantify the safety criticality of each layer's router. Manipulation of only the high-RoSais router(s) can flip the default route into an unsafe one. For instance, on JailbreakBench, masking 5 routers in DeepSeek-V2-Lite increases attack success rate (ASR) by over 4× to 0.79, highlighting an inherent risk that router manipulation may naturally occur in MoE LLMs. We further propose a Fine-grained token-layer-wise Stochastic Optimization framework to discover more concrete Unsafe Routes (F-SOUR), which explicitly considers the sequentiality and dynamics of input tokens. Across four representative MoE LLM families, F-SOUR achieves an average ASR of 0.90 and 0.98 on JailbreakBench and AdvBench, respectively. Finally, we outline defensive perspectives, including safety-aware route disabling and router training, as promising directions to safeguard MoE LLMs. We hope our work can inform future red-teaming and safeguarding of MoE LLMs. Our code is provided in https://github.com/TrustAIRLab/UnsafeMoE.

Access to frontier large language models (LLMs), such as GPT-5 and Gemini-2.5, is often hindered by high pricing, payment barriers, and regional restrictions. These limitations drive the proliferation of shadow APIs, third-party services that claim to provide access to official model services without regional limitations via indirect access. Despite their widespread use, it remains unclear whether shadow APIs deliver outputs consistent with those of the official APIs, raising concerns about the reliability of downstream applications and the validity of research findings that depend on them. In this paper, we present the first systematic audit between official LLM APIs and corresponding shadow APIs. We first identify 17 shadow APIs that have been utilized in 187 academic papers, with the most popular one reaching 5,966 citations and 58,639 GitHub stars by December 6, 2025. Through multidimensional auditing of three representative shadow APIs across utility, safety, and model verification, we uncover both indirect and direct evidence of deception practices in shadow APIs. Specifically, we reveal performance divergence reaching up to 47.21%, significant unpredictability in safety behaviors, and identity verification failures in 45.83% of fingerprint tests. These deceptive practices critically undermine the reproducibility and validity of scientific research, harm the interests of shadow API users, and damage the reputation of official model providers.

The rapid growth of research in LLM safety makes it hard to track all advances. Benchmarks are therefore crucial for capturing key trends and enabling systematic comparisons. Yet, it remains unclear why certain benchmarks gain prominence, and no systematic assessment has been conducted on their academic influence or code quality. This paper fills this gap by presenting the first multi-dimensional evaluation of the influence (based on five metrics) and code quality (based on both automated and human assessment) on LLM safety benchmarks, analyzing 31 benchmarks and 382 non-benchmarks across prompt injection, jailbreak, and hallucination. We find that benchmark papers show no significant advantage in academic influence (e.g., citation count and density) over non-benchmark papers. We uncover a key misalignment: while author prominence correlates with paper influence, neither author prominence nor paper influence shows a significant correlation with code quality. Our results also indicate substantial room for improvement in code and supplementary materials: only 39

Generative foundation models (GenFMs), such as large language models and text-to-image systems, have demonstrated remarkable capabilities in various downstream applications. As they are increasingly deployed in high-stakes applications, assessing their trustworthiness has become both a critical necessity and a substantial challenge. Existing evaluation efforts are fragmented, rapidly outdated, and often lack extensibility across modalities. This raises a fundamental question: how can we systematically, reliably, and continuously assess the trustworthiness of rapidly advancing GenFMs across diverse modalities and use cases? To address these gaps, we introduce TrustGen, a dynamic and modular benchmarking system designed to systematically evaluate the trustworthiness of GenFMs across text-to-image, large language, and vision-language modalities. TrustGen standardizes trust evaluation through a unified taxonomy of over 25 fine-grained dimensions—including truthfulness, safety, fairness, robustness, privacy, and machine ethics—while supporting dynamic data generation and adaptive evaluation through three core modules: Metadata Curator, Test Case Builder, and Contextual Variator. Taking TrustGen into action to evaluate the trustworthiness of 39 models reveals four key insights. (1) State-of-the-art GenFMs achieve promising overall trust performance, yet significant limitations remain in specific dimensions such as hallucination resistance, fairness, and privacy preservation. (2) Contrary to prevailing assumptions, open-source models now rival and occasionally surpass proprietary systems in trustworthiness metrics. (3) The trust gap among top-performing models is narrowing, likely due to increased industry convergence on best practices. (4) Trustworthiness is not an isolated property; it interacts complexly with other behaviors, such as helpfulness and ethical decision-making. TrustGen is a transformative step toward standardized, scalable, and actionable trustworthiness evaluation, supporting dynamic assessments across diverse modalities and trust dimensions that evolve alongside the generative AI landscape.

LLM-powered coding agents increasingly make software supply chain decisions. They generate imports, recommend packages, and write installation commands. Prior work showed that these systems can hallucinate non-existent package names, which attackers may register as malicious packages. In this paper, we show that this risk is not only a passive model failure. It can be actively induced through the persistent Skill artifact. We introduce Dependency Steering, an attack paradigm in which a malicious Skill biases a coding agent toward an attacker-controlled package during benign coding tasks. The attack does not require modifying model weights, training data, or user prompts. To construct realistic attacks, we design a Skill-level optimization method that searches for localized semantic edits that preserve the apparent purpose of the original Skill while increasing targeted package generation. Across multiple coding-oriented LLMs and programming benchmarks, Dependency Steering achieves high targeted hallucination rates, transfers across models and task domains, and remains difficult for evaluated Skill scanners and LLM-based auditors to detect. Our results show that persistent agent instructions form an underexplored software supply chain attack surface.

Prompt learning is a new machine learning paradigm that has attracted ample attention due to its simplicity and proven efficacy. Despite its growing adoption, the security vulnerabilities associated with this paradigm remain underexplored. In this work, we take the first step to propose BadBone, a stealthy and adaptive backdoor attack against prompt learning using bi-level optimization. Instead of backdooring the prompt learning process, we aim to compromise a backbone model such that only target downstream tasks employing prompt learning inherit the backdoor vulnerability. Extensive experiments on three different models and three datasets from various domains show that our targeted/untargeted backdoored models achieve high attack performance while maintaining utility on both pretraining and downstream tasks. Moreover, we evaluate our approach against six state-of-the-art model-level defenses, including Neural Cleanse, ABS, MNTD, NAD, CLP, and D-BR. The results demonstrate that these defenses are largely ineffective against our backdoored models and thus leave the effective defense as an important direction for future work. Our code is available at https://github.com/TrustAIRLab/BadBone.

The rapid development of Language Diffusion Models (LDMs) challenges the dominant position of auto-regressive competitors in language processing. However, their flexible, any-order decoding strategies not only enable fast decoding speed but also potentially bring new trustworthiness challenges. To better understand the risks behind their pipelines, we introduce a comprehensive trustworthiness benchmark tailored to LDMs (TrustLDM), evaluating safety, privacy, and fairness across different LDM architectures with multiple categories of static post contexts. Our empirical results show that although LDMs generally exhibit strong trustworthiness with only the user prompts, their alignment behavior degrades noticeably when the malicious post contexts are attached to the masked responses. We further observe that longer contexts do not necessarily induce stronger effects, and both decoding order and generation length affect the evaluation outcomes. Finally, we propose TrustLDM-Auto, an automatic evaluation framework that leverages LDM decoding flexibility to systematically identify vulnerable configurations, revealing substantial trustworthiness weaknesses across all evaluated models and dimensions. Our work may potentially help the community build more trustworthy LDMs. Our code is available at https://github.com/PKU-ML/TrustLDM.

Transferable adversarial examples raise critical security concerns in real-world, black-box attack scenarios. However, in this work, we identify two main problems in common evaluation practices: (1) For attack transferability, lack of systematic, one-to-one attack comparison and fair hyperparameter settings. (2) For attack stealthiness, simply no comparisons. To address these problems, we establish new evaluation guidelines by (1) proposing a novel attack categorization strategy and conducting systematic and fair intra-category analyses on transferability, and (2) considering diverse imperceptibility metrics and finer-grained stealthiness characteristics from the perspective of attack traceback. To this end, we provide the first large-scale evaluation of transferable adversarial examples on ImageNet, involving 23 representative attacks against 9 representative defenses. Our evaluation leads to a number of new insights, including consensus-challenging ones: (1) Under a fair attack hyperparameter setting, one early attack method, DI, actually outperforms all the follow-up methods. (2) A state-of-the-art defense, DiffPure, actually gives a false sense of (white-box) security since it is indeed largely bypassed by our (black-box) transferable attacks. (3) Even when all attacks are bounded by the same $L_p$ norm, they lead to dramatically different stealthiness performance, which negatively correlates with their transferability performance. Overall, our work demonstrates that existing problematic evaluations have indeed caused misleading conclusions and missing points, and as a result, hindered the assessment of the actual progress in this field.

With the rapid adoption of large language models (LLMs), conversational AI agents have become widely deployed across real-world applications. To enhance safety, these agents are often equipped with guardrails that moderate harmful content. Identifying the guardrails in an agent thus becomes critical for adversaries to understand the system and design guard-specific attacks. In this work, we introduce AP-Test, a novel approach that leverages guard-specific adversarial prompts to detect the identity of guardrails deployed in black-box AI agents. Our method addresses key challenges in this task, including the influence of safety-aligned LLMs and other guardrails, as well as a lack of principled decision-making strategies. AP-Test employs two complementary testing strategies, input and output guard tests, and a new metric, match score, to enable robust identification. Experiments across diverse agents and four open-source guardrails demonstrate that AP-Test achieves perfect classification accuracy in multiple scenarios. Ablation studies further highlight the necessity of our proposed components. Our findings reveal a practical path toward guardrail identification in real-world AI systems.

As academic submissions grow, the traditional peer review process struggles to keep up, raising concerns about quality and fairness. A trend of using large language models (LLMs) for assistance has emerged. In this work, we take a critical step toward improving the quality of LLM-generated reviews. We propose the PeerCheck framework, which investigates LLM-human review differences (RQ1) and explores methods to improve LLM-generated review quality (RQ2). We first analyzed the human-written reviews with reviews generated by various LLMs and found that LLMs and humans focus on different terms, e.g., LLMs prioritize theory while humans emphasize methodology and experiments. We further adopt prompt engineering, such as Chain-of-Thought (CoT), and utilize retrieval-augmented generation (RAG) to enhance the LLM-generated reviews towards human-level quality. We find CoT significantly improves the quality of LLM reviews, while we discover an unexpected "RAG paradox," i.e., experiments with RAG produce different results for various LLMs and, in some cases, even reduce review quality. Our comprehensive analysis of LLM-generated academic reviews illustrates both possibilities and limitations, contributing to a more effective, human-aligned review system. Our dataset is available on https://github.com/TrustAIRLab/PeerCheck.

While large language model-based multi-agent systems have shown strong potential for complex reasoning, how to effectively organize multiple agents remains an open question. In this paper, we introduce OrgAgent, a company-style hierarchical multi-agent framework that separates collaboration into governance, execution, and compliance layers. OrgAgent decomposes multi-agent reasoning into three layers: a governance layer for planning and resource allocation, an execution layer for task solving and review, and a compliance layer for final answer control. By evaluating the framework across reasoning tasks, LLMs, execution modes, and execution policies, we find that multi-agent systems organized in a company-style hierarchy generally outperform other organizational structures. Besides, hierarchical coordination also reduces token consumption relative to flat collaboration in most settings. For example, for GPT-OSS-120B, the hierarchical setting improves performance over flat multi-agent system by 102.73

As Large Language Models (LLMs) are increasingly integrated into academic peer review, their vulnerability to adversarial hidden prompts, i.e., adversarial instructions embedded in submissions to manipulate outcomes, poses a critical threat to scholarly integrity. We propose SafeReview, a co-evolutionary adversarial training framework for defending LLM-based peer review systems against such attacks. SafeReview jointly trains a Generator model to create sophisticated attack prompts and a Defender model to preserve review integrity under adversarial manipulation. The Generator is optimized to produce increasingly effective prompt injections, while the Defender is strengthened through preference-based training to maintain consistent reviews between clean and attacked submissions. Experimental results show that SafeReview improves robustness against adaptive prompt injection attacks, better preserves paper ranking under attack, and generalizes across attacker architectures compared with static defenses. These results demonstrate the potential of co-evolutionary training as a foundation for securing LLM-assisted peer review.

Recent work has applied differential privacy (DP) to adapt large language models (LLMs) for sensitive applications, offering theoretical guarantees. However, its practical effectiveness remains unclear, partly due to LLM pretraining, where overlaps and interdependencies with adaptation data can undermine privacy despite DP efforts. To analyze this issue in practice, we investigate privacy risks under DP adaptations in LLMs using state-of-the-art attacks such as robust membership inference and canary data extraction. We benchmark these risks by systematically varying the adaptation data distribution, from exact overlaps with pretraining data, through in-distribution (IID) cases, to entirely out-of-distribution (OOD) examples. Additionally, we evaluate how different adaptation methods and different privacy regimes impact the vulnerability. Our results show that distribution shifts strongly influence privacy vulnerability: the closer the adaptation data is to the pretraining distribution, the higher the practical privacy risk at the same theoretical guarantee, even without direct data overlap. We find that parameter-efficient fine-tuning methods, such as LoRA, achieve the highest empirical privacy protection for OOD data. Our benchmark identifies key factors for achieving practical privacy in DP LLM adaptation, providing actionable insights for deploying customized models in sensitive settings. Looking forward, we propose a structured framework for holistic privacy assessment beyond adaptation privacy, to identify and evaluate risks across the full pretrain-adapt pipeline of LLMs.

Inference attacks have been widely studied and offer a systematic risk assessment of ML services; however, their implementation and the attack parameters for optimal estimation are challenging for non-experts. The emergence of advanced large language models presents a promising yet largely unexplored opportunity to develop autonomous agents as inference attack experts, helping address this challenge. In this paper, we propose AttackPilot, an autonomous agent capable of independently conducting inference attacks without human intervention. We evaluate it on 20 target services. The evaluation shows that our agent, using GPT-4o, achieves a 100.0

The deployment of large language models (LLMs) raises significant ethical and safety concerns. While LLM alignment techniques are adopted to improve model safety and trustworthiness, adversaries can exploit these techniques to undermine safety for malicious purposes, resulting in misalignment. Misaligned LLMs may be published on open platforms to magnify harm. To address this, additional safety alignment, referred to as realignment, is necessary before deploying untrusted third-party LLMs. This study explores the efficacy of fine-tuning methods in terms of misalignment, realignment, and the effects of their interplay. By evaluating four Supervised Fine-Tuning (SFT) and two Preference Fine-Tuning (PFT) methods across four popular safety-aligned LLMs, we reveal a mechanism asymmetry between attack and defense. While Odds Ratio Preference Optimization (ORPO) is most effective for misalignment, Direct Preference Optimization (DPO) excels in realignment, albeit at the expense of model utility. Additionally, we identify model-specific resistance, residual effects of multi-round adversarial dynamics, and other noteworthy findings. These findings highlight the need for robust safeguards and customized safety alignment strategies to mitigate potential risks in the deployment of LLMs. Our code is available at https://github.com/zhangrui4041/The-Art-of-Mis-alignment.