Pushing the Limits of Safety: A Technical Report on the ATLAS Challenge 2025

Multimodal Large Language Models (MLLMs) have enabled transformative advancements across diverse applications but remain susceptible to safety threats, especially jailbreak attacks that induce harmful outputs. To systematically evaluate and improve their safety, we organized the Adversarial Testing & Large-model Alignment Safety Grand Challenge (ATLAS) 2025}. This technical report presents findings from the competition, which involved 86 teams testing MLLM vulnerabilities via adversarial image-text attacks in two phases: white-box and black-box evaluations. The competition results highlight ongoing challenges in securing MLLMs and provide valuable guidance for developing stronger defense mechanisms. The challenge establishes new benchmarks for MLLM safety evaluation and lays groundwork for advancing safer multimodal AI systems. The code and data for this challenge are openly available at https://github.com/NY1024/ATLAS_Challenge_2025.

AngleRoCL: Angle-Robust Concept Learning for Physically View-Invariant T2I Adversarial Patches

Cutting-edge works have demonstrated that text-to-image (T2I) diffusion models can generate adversarial patches that mislead state-of-the-art object detectors in the physical world, revealing detectors' vulnerabilities and risks. However, these methods neglect the T2I patches' attack effectiveness when observed from different views in the physical world (i.e., angle robustness of the T2I adversarial patches). In this paper, we study the angle robustness of T2I adversarial patches comprehensively, revealing their angle-robust issues, demonstrating that texts affect the angle robustness of generated patches significantly, and task-specific linguistic instructions fail to enhance the angle robustness. Motivated by the studies, we introduce Angle-Robust Concept Learning (AngleRoCL), a simple and flexible approach that learns a generalizable concept (i.e., text embeddings in implementation) representing the capability of generating angle-robust patches. The learned concept can be incorporated into textual prompts and guides T2I models to generate patches with their attack effectiveness inherently resistant to viewpoint variations. Through extensive simulation and physical-world experiments on five SOTA detectors across multiple views, we demonstrate that AngleRoCL significantly enhances the angle robustness of T2I adversarial patches compared to baseline methods. Our patches maintain high attack success rates even under challenging viewing conditions, with over 50% average relative improvement in attack effectiveness across multiple angles. This research advances the understanding of physically angle-robust patches and provides insights into the relationship between textual concepts and physical properties in T2I-generated contents.

Light as Deception: GPT-driven Natural Relighting Against Vision-Language Pre-training Models

While adversarial attacks on vision-and-language pretraining (VLP) models have been explored, generating natural adversarial samples crafted through realistic and semantically meaningful perturbations remains an open challenge. Existing methods, primarily designed for classification tasks, struggle when adapted to VLP models due to their restricted optimization spaces, leading to ineffective attacks or unnatural artifacts. To address this, we propose \textbf{LightD}, a novel framework that generates natural adversarial samples for VLP models via semantically guided relighting. Specifically, LightD leverages ChatGPT to propose context-aware initial lighting parameters and integrates a pretrained relighting model (IC-light) to enable diverse lighting adjustments. LightD expands the optimization space while ensuring perturbations align with scene semantics. Additionally, gradient-based optimization is applied to the reference lighting image to further enhance attack effectiveness while maintaining visual naturalness. The effectiveness and superiority of the proposed LightD have been demonstrated across various VLP models in tasks such as image captioning and visual question answering.

Enhancing Adversarial Robustness of Vision Language Models via Adversarial Mixture Prompt Tuning

Large pre-trained Vision Language Models (VLMs) have excellent generalization capabilities but are highly susceptible to adversarial examples, presenting potential security risks. To improve the robustness of VLMs against adversarial examples, adversarial prompt tuning methods are proposed to align the text feature with the adversarial image feature without changing model parameters. However, when facing various adversarial attacks, a single learnable text prompt has insufficient generalization to align well with all adversarial image features, which finally leads to the overfitting phenomenon. To address the above challenge, in this paper, we empirically find that increasing the number of learned prompts can bring more robustness improvement than a longer prompt. Then we propose an adversarial tuning method named Adversarial Mixture Prompt Tuning (AMPT) to enhance the generalization towards various adversarial attacks for VLMs. AMPT aims to learn mixture text prompts to obtain more robust text features. To further enhance the adaptability, we propose a conditional weight router based on the input adversarial image to predict the mixture weights of multiple learned prompts, which helps obtain sample-specific aggregated text features aligning with different adversarial image features. A series of experiments show that our method can achieve better adversarial robustness than state-of-the-art methods on 11 datasets under different experimental settings.

Fair-PP: A Synthetic Dataset for Aligning LLM with Personalized Preferences of Social Equity

Human preference plays a crucial role in the refinement of large language models (LLMs). However, collecting human preference feedback is costly and most existing datasets neglect the correlation between personalization and preferences. To address this issue, we introduce Fair-PP, a synthetic dataset of personalized preferences targeting social equity, derived from real-world social survey data, which includes 28 social groups, 98 equity topics, and 5 personal preference dimensions. Leveraging GPT-4o-mini, we engage in role-playing based on seven representative persona portrayals guided by existing social survey data, yielding a total of 238,623 preference records. Through Fair-PP, we also contribute (i) An automated framework for generating preference data, along with a more fine-grained dataset of personalized preferences; (ii) analysis of the positioning of the existing mainstream LLMs across five major global regions within the personalized preference space; and (iii) a sample reweighting method for personalized preference alignment, enabling alignment with a target persona while maximizing the divergence from other personas. Empirical experiments show our method outperforms the baselines.

NoiseController: Towards Consistent Multi-view Video Generation via Noise Decomposition and Collaboration

High-quality video generation is crucial for many fields, including the film industry and autonomous driving. However, generating videos with spatiotemporal consistencies remains challenging. Current methods typically utilize attention mechanisms or modify noise to achieve consistent videos, neglecting global spatiotemporal information that could help ensure spatial and temporal consistency during video generation. In this paper, we propose the NoiseController, consisting of Multi-Level Noise Decomposition, Multi-Frame Noise Collaboration, and Joint Denoising, to enhance spatiotemporal consistencies in video generation. In multi-level noise decomposition, we first decompose initial noises into scene-level foreground/background noises, capturing distinct motion properties to model multi-view foreground/background variations. Furthermore, each scene-level noise is further decomposed into individual-level shared and residual components. The shared noise preserves consistency, while the residual component maintains diversity. In multi-frame noise collaboration, we introduce an inter-view spatiotemporal collaboration matrix and an intra-view impact collaboration matrix , which captures mutual cross-view effects and historical cross-frame impacts to enhance video quality. The joint denoising contains two parallel denoising U-Nets to remove each scene-level noise, mutually enhancing video generation. We evaluate our NoiseController on public datasets focusing on video generation and downstream tasks, demonstrating its state-of-the-art performance.

Visibility-Uncertainty-guided 3D Gaussian Inpainting via Scene Conceptional Learning

3D Gaussian Splatting (3DGS) has emerged as a powerful and efficient 3D representation for novel view synthesis. This paper extends 3DGS capabilities to inpainting, where masked objects in a scene are replaced with new contents that blend seamlessly with the surroundings. Unlike 2D image inpainting, 3D Gaussian inpainting (3DGI) is challenging in effectively leveraging complementary visual and semantic cues from multiple input views, as occluded areas in one view may be visible in others. To address this, we propose a method that measures the visibility uncertainties of 3D points across different input views and uses them to guide 3DGI in utilizing complementary visual cues. We also employ uncertainties to learn a semantic concept of scene without the masked object and use a diffusion model to fill masked objects in input images based on the learned concept. Finally, we build a novel 3DGI framework, VISTA, by integrating VISibility-uncerTainty-guided 3DGI with scene conceptuAl learning. VISTA generates high-quality 3DGS models capable of synthesizing artifact-free and naturally inpainted novel views. Furthermore, our approach extends to handling dynamic distractors arising from temporal object changes, enhancing its versatility in diverse scene reconstruction scenarios. We demonstrate the superior performance of our method over state-of-the-art techniques using two challenging datasets: the SPIn-NeRF dataset, featuring 10 diverse static 3D inpainting scenes, and an underwater 3D inpainting dataset derived from UTB180, including fast-moving fish as inpainting targets.

A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment

The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.

Mask Image Watermarking

We present MaskMark, a simple, efficient and flexible framework for image watermarking. MaskMark has two variants: MaskMark-D, which supports global watermark embedding, watermark localization, and local watermark extraction for applications such as tamper detection, and MaskMark-ED, which focuses on local watermark embedding and extraction with enhanced robustness in small regions, enabling localized image protection. Built upon the classical Encoder- Distortion-Decoder training paradigm, MaskMark-D introduces a simple masking mechanism during the decoding stage to support both global and local watermark extraction. A mask is applied to the watermarked image before extraction, allowing the decoder to focus on selected regions and learn local extraction. A localization module is also integrated into the decoder to identify watermark regions during inference, reducing interference from irrelevant content and improving accuracy. MaskMark-ED extends this design by incorporating the mask into the encoding stage as well, guiding the encoder to embed the watermark in designated local regions for enhanced robustness. Comprehensive experiments show that MaskMark achieves state-of-the-art performance in global watermark extraction, local watermark extraction, watermark localization, and multi-watermark embedding. It outperforms all existing baselines, including the recent leading model WAM for local watermarking, while preserving high visual quality of the watermarked images. MaskMark is also flexible, by adjusting the distortion layer, it can adapt to different robustness requirements with just a few steps of fine-tuning. Moreover, our approach is efficient and easy to optimize, requiring only 20 hours on a single A6000 GPU with just 1/15 the computational cost of WAM.

Towards Benchmarking and Assessing the Safety and Robustness of Autonomous Driving on Safety-critical Scenarios

Autonomous driving has made significant progress in both academia and industry, including performance improvements in perception task and the development of end-to-end autonomous driving systems. However, the safety and robustness assessment of autonomous driving has not received sufficient attention. Current evaluations of autonomous driving are typically conducted in natural driving scenarios. However, many accidents often occur in edge cases, also known as safety-critical scenarios. These safety-critical scenarios are difficult to collect, and there is currently no clear definition of what constitutes a safety-critical scenario. In this work, we explore the safety and robustness of autonomous driving in safety-critical scenarios. First, we provide a definition of safety-critical scenarios, including static traffic scenarios such as adversarial attack scenarios and natural distribution shifts, as well as dynamic traffic scenarios such as accident scenarios. Then, we develop an autonomous driving safety testing platform to comprehensively evaluate autonomous driving systems, encompassing not only the assessment of perception modules but also system-level evaluations. Our work systematically constructs a safety verification process for autonomous driving, providing technical support for the industry to establish standardized test framework and reduce risks in real-world road deployment.