KaliBench: A Fine-Grained Benchmark for Cybersecurity Tool Use on Kali Linux with Runtime-Free Verifiable Rewards
KaliBench shows no open-weight model exceeds 42% exact-command accuracy generating Kali Linux CLI commands without tool hints.
KaliBench is a fine-grained benchmark of 8,504 natural-language-to-command pairs covering 1,642 Kali Linux tools, 23 capability dimensions, and five security phases. It uses manuscript-grounded construction, deterministic canonicalization, alias-aware scoring, and verification via LLM checks, sandboxed execution, and human review, and it supplies runtime-free verifiable rewards for training. Across 24 configurations of general-purpose and security-focused open-weight models, none exceeded 42% exact-command accuracy in the unrestricted setting. Supervised fine-tuning and reinforcement learning with KaliBench rewards substantially improved an 8B model to roughly match a 685B mixture-of-experts model.
- 8,504 query-command pairs span 1,642 Kali tools, 23 dimensions, and five phases.
- Verification combines LLM checks, sandboxed execution, and human refinement.
- No open-weight model topped 42% exact-command accuracy without tool hints.
- SFT and verifiable-reward RL lifted an 8B model near a 685B MoE.
Full article227 words · extracted from arxiv.org · click to collapse
LLMs are increasingly applied to cybersecurity workflows, where they are expected to translate analysts' intent into tool invocations. However, existing evaluations focus on knowledge-based assessments or end-to-end agentic tasks, and do not directly measure LLMs' ability to generate executable commands for real-world cybersecurity tools. This gap is critical because cybersecurity operations rely on strict command-line interfaces (CLIs), where minor syntax errors, incorrect flag--value bindings, or argument misordering can invalidate execution. We introduce KaliBench, a fine-grained benchmark and dataset for natural-language--to--CLI translation on Kali Linux, comprising 8,504 query--command pairs spanning 1,642 tools across 23 capability dimensions and 5 security phases. KaliBench is constructed via a manuscript-grounded pipeline with deterministic canonicalization and alias-aware evaluation, enabling precise and reproducible assessment of tool selection and argument construction. To ensure both semantic correctness and practical executability, we develop a multi-stage verification pipeline that combines LLM-based validation, sandboxed terminal execution, and human-in-the-loop refinement. Building on these fine-grained, deterministic signals, KaliBench further enables runtime-free verifiable rewards for training. Across three evaluation modes and 24 configurations of general-purpose and security-focused open-weight models, no open-weight model exceeds 42% exact-command accuracy in the unrestricted setting, highlighting the difficulty of accurate CLI-based cybersecurity tool use without explicit tool hints. We further show that supervised fine-tuning and reinforcement learning with verifiable rewards derived from KaliBench significantly improve an 8B model and achieve performance comparable to a 685B MoE model.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2610.02206