WebMaze steers LLM web scanners into decoy worlds
Original title: WebMaze: Defending Against LLM-Powered Web Scanners with Functionality-Coupled Deception
On 12 applications and three scanners, decoy pages averaged 93.0% of distinct pages requested per run.
The paper studies six LLM-based web scanners and presents WebMaze, a defense that keeps genuine application behavior while sending scanners into generated decoy worlds. Functionality-coupled traps on real pages open view-specific interactive decoys built from existing application views. Under a default no-disclosure setting, decoy pages made up an average of 93.0% of distinct pages requested per run across 12 web applications and three scanners. Scanners still mostly explored decoys after being told about the mechanism and told to avoid them.
- The study covered six LLM-based scanners before designing the defense.
- Traps blend into genuine components and do not change real functionality.
- The reported test used 12 web applications and three scanners.
- Decoy pages averaged 93.0% of distinct pages requested per run.
- Disclosure and avoidance instructions did not stop predominant decoy exploration.
From the original article
1 min read at arxiv.orgLarge language models (LLMs) have enabled more effective vulnerability scanning of web applications. However, these capabilities also make it significantly easier for attackers to automate cyberattacks on the web. In this work, we investigate how to divert malicious LLM-based web scanning, with preservation of genuine functionality as a deployment requirement.
By Pengfei Zeng · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.