arXiv 2610.11942 · cs.CV · 2026-10-08
Right Screen, Wrong Transition: World Models as Verifiers for GUI Agents
Jiaming Zhang, Xuan Wang, Fuyao Zhang, Yang Cao, Lingjuan Lyu, Wei Yang Bryan Lim
机构待核实
在最新榜 #118 本站收录于 2026-10-24
一句话要点
GUI agent 的安全取决于“状态转移”而不只是当前屏幕：点“登录”后出现登录页正常，点“查看订单”后出现登录页就是攻击。LGWM 是不带解码器、按动作条件直接预测下一屏表示的世界模型，用 185 万条真实 GUI 转移训练，验证只需向量比较；在 RSWT-BENCH 上免训练打分 AUC 0.987、每次 17 毫秒，与最强闭源 VLM 相当。
For GUI agents, safety is a property of transitions, not screens: a login page after tapping Sign in is fine, after View order it is an attack. LGWM, a decoder-free action-conditioned world model trained on 1.85M real GUI transitions, predicts the next screen's representation so verification is a vector comparison — 0.987 AUC at 17 ms on RSWT-BENCH, on par with the strongest closed VLMs.
阅读 arXiv 原文 PDF Hugging Face 讨论 代码仓库 项目页
主题与学科
同主题：智能体
- Raven: The Harness of Harnesses for Composable Agentic Intelligence ▲748
- False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents ▲642
- Omni-IO Skills: Harnessing Your Agent Omni-Native ▲485
- BizAgentBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business ▲388
- Player-1: Learning to Play from Synthetic Experience ▲347
同主题：安全与可解释
- Post-Training Leaves Behavioral Shadows on Unrelated Decisions 2026-09-24 · ▲273
- S-Space: Exploring Spatial Workspace in Multimodal Language Models 2026-10-09 · ▲270
- Behavioral Compositionality of Sparse Autoencoder Representations 2026-10-15 · ▲108
- DecepEval: A Benchmark for Evaluating Deception in LLM Agents 2026-10-06 · ▲85
- Last Secure Code Benchmark: Is the Code You Vibe Secure? 2026-10-10 · ▲76
数据说明
元数据来自 arXiv API（已核实）；点赞来自 Hugging Face 每日论文；引用来自 Semantic Scholar。最近刷新 2026-10-26。
要点由本站根据摘要自写，不转载摘要原文，请阅读原文。