CyberFactory：利用真實世界實例擴展網絡安全能力

摘要

隨著大型語言模型（LLMs）在編碼能力上的持續進步，其在網路安全領域的潛力已引起越來越多的研究關注，其中閉源大型語言模型（如 Mythos）提供了先進的網路安全能力。然而，現有的開源努力仍然有限：前沿的開放權重模型未提供可重現的網路安全訓練方案；開源訓練方案聚焦於孤立任務，缺乏可擴展的智慧體數據；而擴展智慧體滾動生成則需要強大的領域先驗。在本研究中，我們提出 CyberFactory，一個統一的開源框架，在概念驗證（PoC）生成、漏洞修補與網路安全問答（CyberQA）之間串連資料建構、軌跡合成與模型訓練。CyberFactory 將公開的漏洞工件（包括來自野外環境的 CVE）轉化為可執行且可驗證的任務實例。它進一步利用一個可重用的漏洞分析技能，引導教師模型進行原始碼檢視、以領域先驗解決問題，以及基於證據的驗證。由此產生的監督是智慧體式的：模型與工具及目標環境互動，並根據執行回饋修訂其解決方案。利用這些軌跡，我們訓練並發布了 Aegis——在希臘神話中，Aegis 是宙斯與雅典娜的防護盾牌；此名稱反映了該模型防禦性、以安全為導向的用途——該模型內化了技能引導的流程，而無需在推論時使用該技能。在 CyberGym 上，Aegis 在一小時預算內達到 52.4% 的 Pass@1，較其 Qwen 3.5 基礎模型提升 22.8 個百分點，並且在相同框架下優於所評估的通用基礎模型。

English

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 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\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.