Skip to content
InfoResearchPreprintLLM-specific

ReconPotBench: Evaluating LLM Honeypots in Multi-Round Linux Reconnaissance

Published
Record updated
View JSON

Summary

ReconPotBench is a benchmark for evaluating LLM-based Linux reconnaissance honeypots across multi-round interactions, using 1,032 three-round interactions in eight categories with reference outputs from a real Linux environment. The study compares Cowrie with open-source and closed-source LLMs on round-level authenticity, cross-round contextual consistency, and hacker-oriented reconnaissance value. It finds that authenticity declines with interaction depth and that LLM honeypots beat Cowrie on authenticity and reconnaissance value, while Cowrie stays more consistent because of its rigid templated behavior.