❌

Normal view

Received — 10 September 2026 ⏭

AgentHijack: Visual Patch Attacks on Multimodal Computer-Use Agents

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zhihao Liu · Hongyu Sun · Zhiyuan Fu · Xiaonan Duan · Jice Wang · Shangru Zhao · Weizhi Meng · Wuxin Yang · Yangfan Zhou · Yuqing Zhang
10 September 2026 at 12:00
arXiv:2609.09212v1 Announce Type: cross Abstract: This paper presents an end-to-end evaluation framework for image-triggered command injection against computer-use agents (CUAs). The goal is to test whether a local visual patch can induce verifiable environmental consequences along the full chain of screenshot input, VLM generation, action parsing, and environment execution. We train and deploy patches on author-controlled GitHub Pages pages and a locally deployed CSDN clone, and evaluate them in real environments across five open-source or publicly available GUI-agent or vision-language-model (VLM) backends. Our experiment aggregates 600 instance-level online cases, with T-ASR, TAPR, and E2E-ASR reaching 84.5%, 47.0%, and 20.3%, respectively. Trajectory analysis further shows that in some successful cases the agent first executes a malicious terminal command and then continues the original benign task. These results indicate that optimized local visual signals can affect not only VLM outputs but also propagate through the execution pipeline of open CUAs and create real environmental risk.
Received — 4 April 2026 ⏭

RAE-AR: Taming Autoregressive Models with Representation Autoencoders

✇cs.AI, q-bio.NC updates on arXiv.org
By: Hu Yu · Hang Xu · Jie Huang · Zeyue Xue · Haoyang Huang · Nan Duan · Feng Zhao
3 April 2026 at 12:00
arXiv:2604.01545v1 Announce Type: new Abstract: The latent space of generative modeling is long dominated by the VAE encoder. The latents from the pretrained representation encoders (e.g., DINO, SigLIP, MAE) are previously considered inappropriate for generative modeling. Recently, RAE method lights the hope and reveals that the representation autoencoder can also achieve competitive performance as the VAE encoder. However, the integration of representation autoencoder into continuous autoregressive (AR) models, remains largely unexplored. In this work, we investigate the challenges of employing high-dimensional representation autoencoders within the AR paradigm, denoted as \textit{RAE-AR}. We focus on the unique properties of AR models and identify two primary hurdles: complex token-wise distribution modeling and the high-dimensionality amplified training-inference gap (exposure bias). To address these, we introduce token simplification via distribution normalization to ease modeling difficulty and improve convergence. Furthermore, we enhance prediction robustness by incorporating Gaussian noise injection during training to mitigate exposure bias. Our empirical results demonstrate that these modifications substantially bridge the performance gap, enabling representation autoencoder to achieve results comparable to traditional VAEs on AR models. This work paves the way for a more unified architecture across visual understanding and generative modeling.
Received — 2 April 2026 ⏭

Microbiome and metabolite signatures for cirrhosis to HCC risk stratification: progress, controversies, and gaps

✇(Multiomics OR Omics) AND (Lung OR gastric OR Hepatocellular)
By: Yanan Duan · Mengting Yang · Miaomiao Li · Yu Sun · Shiguo Liu
1 April 2026 at 18:00

Front Cell Infect Microbiol. 2026 Mar 16;16:1793213. doi: 10.3389/fcimb.2026.1793213. eCollection 2026.

ABSTRACT

The progression from cirrhosis to hepatocellular carcinoma (HCC) is a key outcome in the management of chronic liver disease. This process has a long incubation period and significant individual differences, making early warning still difficult. Clinical follow-up mainly relies on imaging examinations and alpha fetoprotein, but the ability to identify high risk precancerous states is limited. The imbalance of gut microbiota and its metabolites may occur earlier than the visible stage of tumors. They can affect barrier integrity, chronic inflammation, immune surveillance, and metabolic homeostasis through the gut liver axis, and participate in the formation of a pro tumor microenvironment. Therefore, such changes may provide more upstream risk stratification clues for the population with cirrhosis. This article summarizes previous research evidence and summarizes the common microbiome and metabolite characteristics of cirrhosis and high-risk populations, including a decrease in short chain fatty acid (SCFA) related symbiotic bacteria, an increase in inflammation related bacteria, bile acid spectrum shift, and other intestinal derived metabolite abnormalities. This article also outlines the key mechanisms that these features may correspond to, such as barrier damage and microbial translocation, immune suppression, etc. There are still significant uncertainties at present. The effect of SCFA is context dependent. Different etiologies, diets, medications, and complications can lead to significant confounding and affect cross cohort consistency. Subsequent research requires longitudinal cohort validation and the promotion of multi omics integration and the construction of interpretable predictive models to support clinical translation.

PMID:41918873 | PMC:PMC13033666 | DOI:10.3389/fcimb.2026.1793213

Received — 26 March 2026 ⏭

EVA: Efficient Reinforcement Learning for End-to-End Video Agent

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yaolun Zhang · Ruohui Wang · Jiahao Wang · Yepeng Tang · Xuanyu Zheng · Haonan Duan · Hao Lu · Hanming Deng · Lewei Lu
25 March 2026 at 12:00
arXiv:2603.22918v1 Announce Type: cross Abstract: Video understanding with multimodal large language models (MLLMs) remains challenging due to the long token sequences of videos, which contain extensive temporal dependencies and redundant frames. Existing approaches typically treat MLLMs as passive recognizers, processing entire videos or uniformly sampled frames without adaptive reasoning. Recent agent-based methods introduce external tools, yet still depend on manually designed workflows and perception-first strategies, resulting in inefficiency on long videos. We present EVA, an Efficient Reinforcement Learning framework for End-to-End Video Agent, which enables planning-before-perception through iterative summary-plan-action-reflection reasoning. EVA autonomously decides what to watch, when to watch, and how to watch, achieving query-driven and efficient video understanding. To train such agents, we design a simple yet effective three-stage learning pipeline - comprising supervised fine-tuning (SFT), Kahneman-Tversky Optimization (KTO), and Generalized Reward Policy Optimization (GRPO) - that bridges supervised imitation and reinforcement learning. We further construct high-quality datasets for each stage, supporting stable and reproducible training. We evaluate EVA on six video understanding benchmarks, demonstrating its comprehensive capabilities. Compared with existing baselines, EVA achieves a substantial improvement of 6-12% over general MLLM baselines and a further 1-3% gain over prior adaptive agent methods. Our code and model are available at https://github.com/wangruohui/EfficientVideoAgent.
Received — 17 February 2026 ⏭

Automated Proof Generation for Rust Code via Self-Evolution

✇cs.AI, q-bio.NC updates on arXiv.org
By: Tianyu Chen · Shuai Lu · Shan Lu · Yeyun Gong · Chenyuan Yang · Xuheng Li · Md Rakib Hossain Misu · Hao Yu · Nan Duan · Peng Cheng · Fan Yang · Shuvendu K Lahiri · Tao Xie · Lidong Zhou
17 February 2026 at 13:00
arXiv:2410.15756v3 Announce Type: replace-cross Abstract: Ensuring correctness is crucial for code generation. Formal verification offers a definitive assurance of correctness, but demands substantial human effort in proof construction and hence raises a pressing need for automation. The primary obstacle lies in the severe lack of data-there is much fewer proofs than code snippets for Large Language Models (LLMs) to train upon. In this paper, we introduce SAFE, a framework that overcomes the lack of human-written proofs to enable automated proof generation of Rust code. SAFE establishes a self-evolving cycle where data synthesis and fine-tuning collaborate to enhance the model capability, leveraging the definitive power of a symbolic verifier in telling correct proofs from incorrect ones. SAFE also re-purposes the large number of synthesized incorrect proofs to train the self-debugging capability of the fine-tuned models, empowering them to fix incorrect proofs based on the verifier's feedback. SAFE demonstrates superior efficiency and precision compared to GPT-4o. Through tens of thousands of synthesized proofs and the self-debugging mechanism, we improve the capability of open-source models, initially unacquainted with formal verification, to automatically write proofs for Rust code. This advancement leads to a significant improvement in performance, achieving a 52.52% accuracy rate in a benchmark crafted by human experts, a significant leap over GPT-4o's performance of 14.39%.
There are no more articles
❌