Uncensored and Offensive Security AI Models Benchmark
🇬🇧 English
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
🇸🇦 العربية
نماذج الذكاء الاصطناعي للأمن الهجومي: نماذج LLMs للأمن السيبراني غير الخاضعة للرقابة
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
هل هذه النماذج مناسبة لعمليات الأمن في الإنتاج؟
هذه النماذج مخصصة للاختراق المصرح به واختبار الاختراق والبحث الأمني. يجب على المستخدمين التأكد من الامتثال للقوانين المحلية والحصول على التفويض المناسب قبل النشر. هذه النماذج غير خاضعة للرقابة وقد تولد محتوى مهين؛ مطلوب الاستخدام المسؤول.
🇧🇩 বাংলা
আক্রমণাত্মক নিরাপত্তা AI মডেল: সেন্সর-মুক্ত সাইবার নিরাপত্তা LLMs
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
ক्या এই মডেলগুলি উৎপাদন নিরাপত্তা অপারেশনের জন্য উপযুক্ত?
এই মডেলগুলি অনুমোদিত রেড টিমিং, পেনেট্রেশন টেস্টিং এবং নিরাপত্তা গবেষণের জন্য ডিজাইন করা হয়েছে। ব্যবহারকারীদের স্থানীয় আইন মেনে চলতে হবে এবং বাস্তবায়ন করার আগে সঠিক অনুমতি পেতে হবে। এই মডেলগুলি সেন্সর-মুক্ত এবং আক্রমণামূলক বিষয়বস্তু তৈরি করতে পারে; দায়বদ্ধ ব্যবহার প্রয়োজন।
🇩🇪 Deutsch
Offensive Sicherheits-KI-Modelle: Unzensurierte Cybersicherheits-LLMs
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
Sind diese Modelle für produktive Sicherheitsoperationen geeignet?
Diese Modelle sind für autorisiertes Red Teaming, Penetrationstests und Sicherheitsforschung vorgesehen. Benutzer müssen die Einhaltung lokaler Gesetze sicherstellen und vor der Bereitstellung eine angemessene Genehmigung einholen. Diese Modelle sind unzensuriert und können anstößige Inhalte erzeugen; verantwortungsvoller Einsatz ist erforderlich.
🇪🇸 Español
Modelos de IA de Seguridad Ofensiva: LLMs de Ciberseguridad Sin Censura
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
¿Son estos modelos adecuados para operaciones de seguridad en producción?
Estos modelos están diseñados para pruebas de penetración autorizadas, equipo rojo y investigación de seguridad. Los usuarios deben asegurar el cumplimiento de las leyes locales y obtener la autorización adecuada antes de la implementación. Estos modelos son sin censura y pueden generar contenido ofensivo; se requiere un uso responsable.
🇫🇷 Français
Modèles d'IA de Sécurité Offensive : LLMs de Cybersécurité Non Censurés
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
Ces modèles sont-ils adaptés aux opérations de sécurité en production ?
Ces modèles sont destinés aux tests d'intrusion autorisés, au red teaming et à la recherche en sécurité. Les utilisateurs doivent s'assurer de la conformité avec les lois locales et obtenir une autorisation appropriée avant le déploiement. Ces modèles sont non censurés et peuvent générer du contenu offensif ; une utilisation responsable est requise.
🇮🇳 हिन्दी
आक्रामक सुरक्षा AI मॉडल: अनसेंसर्ड साइबर सुरक्षा LLMs
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
क्या ये मॉडल उत्पादन सुरक्षा संचालन के लिए उपयुक्त हैं?
ये मॉडल अधिकृत रेड टीमिंग, प्रवेश परीक्षण और सुरक्षा अनुसंधान के लिए предназначены हैं। उपयोगकर्ताओं को स्थानीय कानूनों का पालन सुनिश्चित करना चाहिए और तैनाती से पहले उचित प्राधिकरण प्राप्त करना चाहिए। ये मॉडल अनसेंसर्ड हैं और अपमानजनक सामग्री उत्पन्न कर सकते हैं; जिम्मेदार उपयोग आवश्यक है।
🇮🇩 Bahasa Indonesia
Model AI Keamanan Ofensif: LLMs Keamanan Siber Tanpa Sensor
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
Apakah model-model ini cocok untuk operasi keamanan produksi?
Model-model ini dimaksudkan untuk red teaming yang diotorisasi, pengujian penetrasi, dan penelitian keamanan. Pengguna harus memastika kepatuhan terhadap hukum setempat dan mendapatkan otorisasi yang tepat sebelum penerapan. Model-model ini tidak ter sensor dan dapat menghasilkan konten ofensif; diperlukan penggunaan yang bertanggung jawab.
🇯🇵 日本語
攻撃的セキュリティAIモデル：検閲なしのサイバーセキュリティLLMs
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
これらのモデルは本番環境のセキュリティ運用に適していますか？
これらのモデルは、認可されたレッドチーム、ペネトレーションテスト、およびセキュリティリサーチのために設計されています。ユーザーは、現地の法律を遵守し、展開前に適切な認可を取得する必要があります。これらのモデルは検閲なしであり、攻撃的なコンテンツを生成する可能性があります；責任ある使用が必要です。
🇧🇷 Português
Modelos de IA de Segurança Ofensiva: LLMs de Cibersegurança Sem Censura
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
Esses modelos são adequados para operações de segurança em produção?
Esses modelos são destinados a testes de invasão autorizados, equipe vermelha e pesquisa de segurança. Os usuários devem garantir a conformidade com as leis locais e obter a devida autorização antes da implantação. Esses modelos são sem censura e podem gerar conteúdo ofensivo; é necessário uso responsável.
🇷🇺 Русский
Наступательные модели ИИ в безопасности: нецензурированные LLMs для кибербезопасности
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
Подходят ли эти модели для эксплуатационных операций безопасности?
Эти модели предназначены для авторизованного красного팀инга, тестирования на проникновение и исследований в области безопасности. Пользователи должны обеспечить соблюдение местных законов и получить соответствующее разрешение перед развертыванием. Эти модели нецензурированы и могут генерировать оскорбительный контент; требуется ответственное использование.
🇨🇳 简体中文
进攻性安全AI模型：未经审查的网络安全LLMs
A curated list of open-weight, uncensored AI models designed for authorized red teaming, penetration testing, and security research. All entries are sourced from Hugging Face model cards and official publications as of September 2026. The repository provides technical specifications, training methodologies, and download links for each model.
The collection includes Deep Hat V2 (White Rabbit Neo), a 7B/32B model trained on 1.7M security samples from USENIX Security 2024. Bug Trace AI-CORE-Apex (26B) and Bug Trace AI-CORE-Ultra (27B) leverage Hacker One data and CVE disclosures for offensive tooling generation. CYBER-FROST-3.8 (Blackfrost-AI) offers ~180B parameters with multi-GPU support and proprietary security corpus training.
Defensive-focused options include Cyber Pal 2.0 (20B) for CTI and SOC workflows, and Cyber-Prime 1.1 (2.6B) with RL-guided post-training on 75K cybersecurity rows. Cyber-Ornith-1.5-9B (Duo Neural) specializes in agentic cybersecurity with function-calling and terminal automation. All models specify VRAM requirements, context lengths, licenses, and Hugging Face download URLs.
这些模型适用于生产安全操作吗？
这些模型旨在用于授权的红队渗透测试和安全研究。用户必须确保符合当地法律并在部署前获得适当授权。这些模型未经审查，可能生成冒犯性内容；需要负责任地使用。