arXiv cs.AI· Sarim Hashmi, Mukul Ranjan, Kshitij Mishra, Mikhail Kuznetsov, Praneeth Vepakomma, Nils Lukas·· 10 天前AI 评分24
AdvSim2Real：在 Web 世界模型中用自适应提示注入训练 Web 智能体
AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model
AI 导读
AdvSim2Real 在冻结的 Web 世界模型中协同演化任务课程、注入对抗者和智能体，训练出更强大也更鲁棒的 4B Web 智能体。该智能体在从未训练过的前沿模型对抗下仍保持完成率，能力提升还可迁移到真实浏览器；在 150 个 Web 任务上，面对这一未见过的对抗者，完成率相对基线智能体提升 33.6%。
正文 · 原文
Abstract:Web agents complete user requests by reading and acting on pages that third parties write, so an instruction planted on a page can redirect the agent away from the user's goal. The agent cannot simply ignore the page, because the page also holds the values and controls the task requires. Current defenses fine-tune the agent on injections fixed before training, and attackers that adapt to the trained model bypass them. Adversarial training lets the attacker adapt but keeps the tasks fixed, so a task stops teaching once the agent solves it. We introduce AdvSim2Real, which co-evolves a task curriculum, an injection adversary, and the agent inside a frozen web world model. The curriculum is rewarded for tasks the agent solves about half of the time, and the adversary only for a success flip, an injection that turns a judged success into a failure. Training in the simulator makes a 4B agent both more capable and more robust: its completion rises with and without attacks, holds against a frontier-model adversary it never trained against, and its capability gain carries over to a real browser. On 150 web tasks, AdvSim2Real raises completion under this unseen adversary by 33.6\% relative to the base agent.
| Comments: | Code at this https URL |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.08773 [cs.CL] |
| (or arXiv:2610.08773v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.08773 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mukul Ranjan [view email]
[v1]
Tue, 6 Oct 2026 17:56:43 UTC (360 KB)
来源：arXiv cs.AI · arxiv.org