Breaking Darknet CAPTCHAs with general purpose LLM
Cryptography and SecurityComputer Vision and Pattern Recognition
Summary
The authors studied how well automated tools can solve special types of CAPTCHAs often used on the darknet that do not use JavaScript. They tested three kinds of challenges that require finding or aligning shapes or choosing objects. They found that while big language and vision models can recognize important features, they have trouble with details like exact placement and rotation. To fix this, the authors combined these models with traditional image processing methods, letting the models handle reasoning and the algorithms handle geometry. This combined approach worked much better, solving over 90% of the challenges.
Authors
Benjamin Fehrensen, Jens Hubler
Abstract
Our work evaluates the effectiveness of automated methods for solving CAPTCHA challenges commonly encountered in darknet environments. These CAPTCHAs are typically designed to operate without JavaScript, resulting in distinct characteristics compared to mainstream CAPTCHA systems. Our study considers three representative challenge types: open-circle localization, rotation-based alignment, and object-selection CAPTCHAs. The experiments reveal a systematic limitation of contemporary MLLMs: while they are generally capable of identifying relevant visual structures, they frequently struggle with precise spatial localization and geometric transformations. These deficiencies can be mitigated either through task reformulation or by augmenting the models with specialized image processing tools. These deficiencies can be mitigated by task reformulation or by equipping the model with specialized image-processing tools. We therefore propose a hybrid framework in which an MLLM serves as a high-level reasoning and orchestration layer while delegating geometric computations to deterministic algorithms via the Model Context Protocol (MCP). The resulting system achieves success rates above 90% across all evaluated CAPTCHA types and demonstrates that combining the complementary strengths of MLLMs and classical computer vision yields a more accurate and efficient solver than either approach alone.