Harness Matters: Measuring and Dissecting Security Resilience in Web Agents
cs.CR, cs.AI
Submitted: 2026-11-09
Updated: 2026-11-09
Code: https://github.com/Agrfeblve/WebTrapFarm
Terminology
Sources
- SafeArena: Evaluating the Safety of Autonomous Web Agents
- VPI-Bench: Visual Prompt Injection Attacks for Computer-Use Agents
- EIA: Environmental Injection Attack on Generalist Web Agents for Privacy Leakage
- AdvAgent: Controllable Blackbox Red-teaming on Web Agents
- When Bots Take the Bait: Exposing and Mitigating the Emerging Social Engineering Attack in Web Automation Agent
- Investigating the Impact of Dark Patterns on LLM-Based Web Agents
- BrowseSafe: Understanding and Preventing Prompt Injection Within AI Browser Agents
- Systems Security Foundations for Agentic Computing
- Mind the Web: The Security of Web Use Agents
- In-Browser LLM-Guided Fuzzing for Real-Time Prompt Injection Testing in Agentic AI Browsers
- WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models
- MindSearch: Mimicking Human Minds Elicits Deep AI Searcher
- The BrowserGym Ecosystem for Web Agent Research
- Dissecting Adversarial Robustness of Multimodal LM Agents
- Prompt Injection attack against LLM-integrated Applications
- Toxicity in ChatGPT: Analyzing Persona-assigned Language Models
- Role-Play with Large Language Models
- Jailbreaking ChatGPT via Prompt Engineering: An Empirical Study
- Manipulating LLM Web Agents with Indirect Prompt Injection Attack via HTML Accessibility Tree
- Ignore Previous Prompt: Attack Techniques For Language Models
Related papers
- From Strategy Search to Policy Learning: MCTS-Guided Adversarial Co-Evolution for LLM Safety Alignment
- Privacy Leakage from Recurrent State in LLM Inference
- Incrementally Verifiable Oracle Aided Computation
- Privilege and confidentiality in generative AI workflows
- BadMoE: Backdooring Mixture-of-Experts LLMs via Routing-Aware Triggers and Task-Underutilized Experts
- CoVault: Secure, Scalable Analytics of Personal Data