Halligan

Introduction This repository provides the official artifact for the paper:"Are CAPTCHAs Still Bot-Hard? Generalized Visual CAPTCHA Solving with Agentic Vision Language Models" Our work explores the effectiveness of Vision-Language Model (VLM) agents in solving modern visual CAPTCHAs by leveraging reasoning, abstraction, and code synthesis capabilities. Contents benchmark.zip: An interactive offline benchmark suite designed to evaluate VLM agents on their ability to solve visual CAPTCHA challenges. halligan.zip: The implementation of Halligan, our proposed VLM agent introduced in the paper