ML Security PapersSnapGuard: Lightweight Prompt Injection Detection for Screenshot-Based Web Agents
Mengyao Du 1, Han Fang 2, Haokai Ma 3, Jiahao Chen 4, Kai Xu 1, Quanjun Yin 1, Ee-Chien Chang 3
1 National University of Defense Technology
2 University of Science and Technology of China
Published on arXiv
2604.25562
Prompt Injection
OWASP LLM Top 10 — LLM01
Excessive Agency
OWASP LLM Top 10 — LLM08
Key Finding
Achieves F1 score of 0.75 across eight attack types, outperforming GPT-4o-prompt while being 8x faster (1.81s vs 14.50s) with zero memory overhead
SnapGuard
Novel technique introduced
Web agents have emerged as an effective paradigm for automating interactions with complex web environments, yet remain vulnerable to prompt injection attacks that embed malicious instructions into webpage content to induce unintended actions. This threat is further amplified for screenshot-based web agents, which operate on rendered visual webpages rather than structured textual representations, making predominant text-centric defenses ineffective. Although multimodal detection methods have been explored, they often rely on large vision-language models (VLMs), incurring significant computational overhead. The bottleneck lies in the complexity of modern webpages: VLMs must comprehend the global semantics of an entire page, resulting in substantial inference time and GPU memory usage. This raises a critical question: can we detect prompt injection attacks from screenshots in a lightweight manner? In this paper, we observe that injected webpages exhibit distinct characteristics compared to benign ones from both visual and textual perspectives. Building on this insight, we propose SnapGuard, a lightweight yet accurate method that reformulates prompt injection detection as multimodal representation analysis over webpage screenshots. SnapGuard leverages two complementary signals: a visual stability indicator that identifies abnormally smooth gradient distributions induced by malicious content, and action-oriented textual signals recovered via contrast-polarity reversal. Extensive evaluations across eight attacks and two benign settings demonstrate that SnapGuard achieves an F1 score of 0.75, outperforming GPT-4o-prompt while being 8x faster (1.81s vs. 14.50s) and introducing no additional memory overhead.
Key Contributions
- Reformulates prompt injection detection as multimodal representation analysis over webpage screenshots rather than structured text
- Combines visual stability indicator (detecting smooth gradient distributions from malicious content) with action-oriented textual signals via contrast-polarity reversal
- Achieves 8x faster inference than GPT-4o baseline (1.81s vs 14.50s) with no additional memory overhead while maintaining F1 of 0.75