--- language: - en - tr - zh pipeline_tag: text-generation tags: - cybersecurity - security - infosec - ai-security - agent - ai-agent - agentic-ai - llm - cs-eval - benchmark - leaderboard - evaluation - source-grounded - critical-infrastructure ---

Garrulus-Agent logo: a European roller on an oak branch

# Garrulus-Agent **Source-Grounded Cybersecurity AI Agent** [![LinkedIn](https://img.shields.io/badge/LinkedIn-Emrah%20D%C3%BCndar-0A66C2)](https://www.linkedin.com/in/emrahdundar/) [![Website](https://img.shields.io/badge/Website-garrulus.com.tr-10203F)](https://garrulus.com.tr) [![CS-Eval](https://img.shields.io/badge/CS--Eval-%231%20overall%20%C2%B7%2099.38-C97A3E)](https://cs-eval.com/#/app/leaderBoard) [![GitHub](https://img.shields.io/badge/GitHub-garrulus--agent-181717)](https://github.com/garrulus-agent) Garrulus-Agent is an AI agent for cybersecurity knowledge that checks its answers against trusted sources. It is an independent project, built by [Emrah Dündar](https://www.linkedin.com/in/emrahdundar/) on his own initiative. ## CS-Eval result - **#1 overall on CS-Eval with a score of 99.38** (as of November 2026) - The first model on CS-Eval to score a perfect 100 in 6 categories - The first model on CS-Eval to cross the 99 mark overall CS-Eval is an open evaluation of cybersecurity knowledge for large language models: 4,369 questions across 11 domains and 42 subdomains. | Category | Score | |---|---| | System and software security fundamentals | **100** | | Cryptography and key management | **100** | | AI & Cybersecurity | **100** | | Business continuity and emergency response | **100** | | Security architecture design | **100** | | Supply chain security | **100** | | Data security and privacy | 99.60 | | Infrastructure security | 99.17 | | Threat detection and prevention | 99.08 | | Access control and identity management | 98.85 | | Vulnerability management and penetration testing | 98.54 | | **Overall** | **99.38** | Source: [official CS-Eval leaderboard](https://cs-eval.com/#/app/leaderBoard). ## Evidence Screenshots of the official CS-Eval leaderboard (cs-eval.com, 12 November 2026). Images are cropped, data is unchanged. **Overall ranking** ![CS-Eval overall ranking: Garrulus-Agent first with 99.38](https://huggingface.co/emrahdundar/Garrulus-Agent/resolve/main/garrulus-agent-cs-eval-overall-ranking.webp) **Category ranking** ![CS-Eval category ranking: Garrulus-Agent category results](https://huggingface.co/emrahdundar/Garrulus-Agent/resolve/main/garrulus-agent-cs-eval-category-ranking.webp) More details: [garrulus.com.tr](https://garrulus.com.tr/en/). ## Availability Model weights are not released at this time. This page is the official Hugging Face home of the project. ## About the author **Emrah Dündar** is an IT and cybersecurity leader with more than 19 years in critical infrastructure security, working on AI strategy and applied AI. Open to conversations on AI strategy, applied AI and cybersecurity leadership. **[Connect on LinkedIn](https://www.linkedin.com/in/emrahdundar/)**. ## Links - Results and evidence: https://garrulus.com.tr - Official leaderboard: https://cs-eval.com/#/app/leaderBoard - GitHub: https://github.com/garrulus-agent - Contact: info@garrulus.com.tr --- **Türkçe:** Garrulus-Agent, cevaplarını güvenilir kaynaklarla kontrol eden bir siber güvenlik yapay zekâ ajanıdır. CS-Eval genel sıralamasında 99,38 puanla birinci, altı kategoride 100 tam puan alan ve genel puanda 99 eşiğini geçen ilk model. Sonuçlar ve kanıt: https://garrulus.com.tr **中文:** Garrulus-Agent 是一款网络安全知识 AI 智能体,会依据可信来源核验自己的答案。它在 CS-Eval 综合排名中以 99.38 分位列第一,是首个在 6 个类别中获得 100 满分、也是首个综合得分突破 99 的模型。结果与证据:https://garrulus.com.tr/en/