ArXiv Releases CyberFactory: A Unified Open-Source Framework for Scaling Cybersecurity Capabilities

Summary:The ArXiv team introduces CyberFactory, a unified open-source framework designed to address the limitations of existing cybersecurity training solutions. By connecting data construction, trajectory synthesis, and model training, CyberFactory provides executable task instances for proof-of-concept (POC) generation, vulnerability patching, and cybersecurity question answering (CyberQA). It transforms public vulnerability artifacts, such as CVEs, into verifiable tasks and employs a reusable vulnera

---

Key Breakthroughs

The ArXiv team has introduced CyberFactory, a unified open-source framework aimed at addressing the limitations of current cybersecurity training solutions. Key features of the framework include:

• Unified Framework Design: CyberFactory integrates data construction, trajectory synthesis, and model training, providing an end-to-end solution from proof-of-concept (POC) generation to cybersecurity question answering (CyberQA).
• Executable Task Instances: The framework transforms public vulnerability artifacts, such as CVEs, into executable and verifiable task instances, ensuring the effectiveness and diversity of training data.
• Reusable Skill Guidance: By introducing a reusable vulnerability-analysis skill, CyberFactory guides the training process and ensures the model does not rely on the skill during inference.
• Model Release: The trained model, Aegis, demonstrates strong performance in the CyberGym benchmark, showcasing its potential in cybersecurity applications.

Technical Highlights

1. Data Construction and Trajectory Synthesis: CyberFactory addresses the issue of insufficient training data by converting public vulnerability artifacts into executable task instances.
2. Reusable Skill Guidance: The framework employs a reusable vulnerability-analysis skill to guide the training process and enhance model performance.
3. Model Training and Release: The Aegis model, trained using CyberFactory, exhibits excellent performance in the CyberGym benchmark, highlighting its capabilities in handling complex cybersecurity tasks.

Industry Impact

The release of CyberFactory brings new technological pathways to the cybersecurity field, with significant implications in the following areas:

• Enhanced Training Efficiency: By providing executable task instances and reusable skill guidance, CyberFactory significantly improves the efficiency of cybersecurity training.
• Improved Model Performance: The strong performance of the Aegis model in the CyberGym benchmark demonstrates its potential in tackling complex cybersecurity challenges.
• Promotion of Open Source Development: As an open-source framework, CyberFactory provides a powerful tool for researchers and developers, fostering open collaboration and innovation in cybersecurity technologies.

Recommendations for Developers

• Explore Framework Features: Developers are encouraged to thoroughly understand the features of CyberFactory and apply it to different cybersecurity training scenarios.
• Contribute to the Open Source Community: Developers are invited to contribute code and resources to CyberFactory, collectively advancing the development of the framework.
• Evaluate Model Performance: Assess the performance of the Aegis model in various benchmarks and optimize and adjust it according to specific needs.

---

Source: ArXiv cs.CL (2026-08-24)

— END —

Tags: #ArXiv #Cybersecurity #Open-Source Framework #Model Training #CyberFactory