Do Agents Dream of Root Shells? Partial-Credit Evaluation of LLM Agents in Capture the Flag Challenges

TL;DR

This paper introduces DeepRed, a benchmark for evaluating LLM agents in realistic cybersecurity Capture The Flag challenges, highlighting current limitations in agent performance and capabilities.

Contribution

The paper presents DeepRed, an open-source benchmark with a novel partial-credit scoring method for assessing LLM agents in complex cybersecurity tasks.

Findings

Best model achieves 35% average checkpoint completion

Agents perform well on common challenges but struggle with discovery tasks

Current LLM agents have limited capabilities in realistic CTF scenarios

Abstract

Large Language Model (LLM) agents are increasingly proposed for autonomous cybersecurity tasks, but their capabilities in realistic offensive settings remain poorly understood. We present DeepRed, an open-source benchmark for evaluating LLM-based agents on realistic Capture The Flag (CTF) challenges in isolated virtualized environments. DeepRed places an agent in a Kali attacker environment with terminal tools and optional web search, connected over a private network to a target challenge, and records full execution traces for analysis. To move beyond binary solved/unsolved outcomes, we introduce a partial-credit scoring method based on challenge-specific checkpoints derived from public writeups, together with an automated summarise-then-judge labelling pipeline for assigning checkpoint completion from logs. Using DeepRed, we benchmark ten commercially accessible LLMs on ten VM-based CTF challenges spanning different challenge categories. The results indicate that current agents remain limited: the best model achieves only 35% average checkpoint completion, performing strongest on common challenge types and weakest on tasks requiring non-standard discovery and longer-horizon adaptation.