Search papers, labs, and topics across Lattice.
This paper introduces CyberFactory, an open-source framework that integrates data construction, trajectory synthesis, and model training to enhance cybersecurity capabilities using large language models. By transforming public vulnerability artifacts into executable task instances and employing a reusable vulnerability-analysis skill, CyberFactory enables agentic interactions where models can revise solutions based on execution feedback. The resulting model, Aegis, achieves a 52.4% Pass@1 score on CyberGym, significantly outperforming its base model and other general-purpose models in the same environment.
Aegis, trained with CyberFactory, outperforms existing models by 22.8 points in cybersecurity tasks, showcasing the power of agentic learning in real-world applications.
As large language models (LLMs) continue to advance in coding capabilities, their potential in cybersecurity has drawn increasing research attention, with closed-source LLMs (e.g., Mythos) delivering advanced cybersecurity capabilities. However, existing open-source efforts remain limited: frontier open-weight models do not provide reproducible cybersecurity training solutions, open-source training solutions focus on isolated tasks and lack scalable agentic data, and scaling agentic rollouts requires strong domain priors. In this work, we introduce \textbf{CyberFactory}, a unified open-source framework that connects data construction, trajectory synthesis, and model training across proof-of-concept (PoC) generation, vulnerability patching, and cybersecurity question answering (CyberQA). CyberFactory transforms public vulnerability artifacts, including CVEs from the wild, into executable and verifiable task instances. It further uses a reusable vulnerability-analysis skill to guide the teacher through source inspection, problem solving with domain prior, and evidence-based validation. The resulting supervision is agentic: the model interacts with tools and target environments and revises its solutions according to execution feedback. Using these trajectories, we train and release \modelname\footnote{\emph{Aegis} is, in Greek mythology, the protective shield of Zeus and Athena; the name reflects the model's defensive, security-oriented purpose.}, which internalizes the skill-guided procedure without requiring the skill at inference time. On CyberGym, \modelname reaches 52.4% Pass@1 under a one-hour budget, improving over its Qwen~3.5 base model by +22.8 points and outperforming the evaluated general-purpose backbones under the same scaffold.
