PopupAttack — security grade SAFE, quality 64/100

Security audit verdict: SAFE · quality 64/100

No red flags found in any of the 11 categories — no credential harvesting, no data exfiltration, no curl-pipe-shell installer. Scanned against the SlowMist agent-security taxonomy, refreshed every 8 hours. Full audit →

by SALT-NLP · Agent Tool · ★ 52

Last updated: · Indexed by AgentSkillsHub · Auto-synced every 8h

🔒 Is PopupAttack safe to install? View the security audit →

About PopupAttack

Attacking Vision-Language Computer Agents via Pop-ups Yanzhe Zhang, Tao Yu, Diyi Yang Overview Autonomous agents powered by large vision and language models (VLM) have demonstrated significant potential in completing daily computer tasks, such as browsing the web to book travel and operating desktop software, which requires agents to understand these interfaces. Despite such visual inputs becoming more integrated into agentic applications, what types of risks and attacks exist around them still remain unclear. In this work, we demonstrate that VLM agents can be easily attacked by a set of carefully designed adversarial pop-ups, which human users would typically recognize and ignore. This distraction leads agents to click these pop-ups instead of performing the tasks as usual. Integrating these pop-ups into existing agent testing environments like OSWorld and VisualWebArena leads to an attack success rate (the frequency of the agent clicking the pop-ups) of 86% on average and decreases the task success rate by 47%. Basic defense techniques such as asking the agent to ignore pop-ups or including an advertisement notice, are ineffective against the attack.

attackclaude-3-5-sonnetcomputer-usellm-agentpop-upvision-language-model

Quick Facts

Stars52
Forks3
LanguagePython
CategoryAgent Tool
Quality Score64.3345722823544/100
Open Issues1
Last Updated2024-12-23
Created2024-11-04
Platformsclaude-code, python
Est. Tokens~13278k

PopupAttack alternative? Top 6 similar tools

Looking for a PopupAttack alternative? If you're comparing PopupAttack with other agent tool tools, these 6 projects are the closest alternatives on Agent Skills Hub — ranked by topic overlap, star count, and community traction.

  • Auto-Use by auto-use · ⭐ 120

    Auto-Use Computer Use — drives your OS, browser, scours the web, writes your code. One agent, end to end.

  • EdgeBox by BIGPPWONG · ⭐ 212

    A fully-featured, GUI-powered local LLM Agent sandbox with complete MCP protocol support. Features both CLI

  • NEWTON by CUTEPKQ · ⭐ 143

    NEWTON: Agentic Planning for Physically Grounded Video Generation

  • wechat-mac-rpa by wq19901103wq · ⭐ 107

    基于多模态视觉感知与 LLM Agent 的 macOS 微信自动化框架 | Visual RPA for WeChat

  • VideoGLaMM by mbzuai-oryx · ⭐ 105

    [CVPR 2025 🔥]A Large Multimodal Model for Pixel-Level Visual Grounding in Videos

  • OmniAgent by HarryHsing · ⭐ 73

    OmniAgent (ICML 2026): the first native omni-modal agent for active video perception — a 7B agent that beats Q

More Agent Tool Tools

Explore other popular agent tool tools:

View all Agent Tool tools →

Popular Python Agent Tools

Frequently Asked Questions

What is PopupAttack?

PopupAttack is Code repo for the paper: Attacking Vision-Language Computer Agents via Pop-ups. It is categorized as a Agent Tool with 52 GitHub stars.

What programming language is PopupAttack written in?

PopupAttack is primarily written in Python. It covers topics such as attack, claude-3-5-sonnet, computer-use.

How do I install or use PopupAttack?

You can find installation instructions and usage details in the PopupAttack GitHub repository at github.com/SALT-NLP/PopupAttack. The project has 52 stars and 3 forks, indicating an active community.

What are the best alternatives to PopupAttack?

The top alternatives to PopupAttack on Agent Skills Hub include Auto-Use, EdgeBox, NEWTON. Each offers a different approach to the same problem space — compare them side-by-side by stars, quality score, and community activity.

How this security grade is produced

Grades come from a rule-based scan built on the SlowMist agent-security taxonomy, covering 11 red-flag categories including credential harvesting, data exfiltration, and curl | sh installers. It is a first-layer scan, not a manual audit — we say so rather than overstate it.

The scale of the problem is documented independently: Liu et al. (2026), in a study of 31,132 agent skills, report that 26.1% contain security vulnerabilities. Our own full-catalog census is published as a citable open dataset.

Sources & who's responsible:

View on GitHub → Browse Agent Tool tools