360 如何训练图龙凤，使其在 CyberGym 漏洞基准测试中名列前茅

发生了什么

据量子位报道，360 的图龙凤安全系统在 CyberGym 上实现了 99.4% 的漏洞复现率，这一数据基于三次独立评测，平均值为 99.38%。文章称，360 的方法是将使用安全数据和专家经验训练的安全专用模型，与能够分析代码、使用工具、生成 PoC 并验证漏洞的专用智能体相结合。文章还讲述了系统新增的一条规则：此前，沙箱曾阻止它使用一个未指明的测试地址。

接下来读什么

• 今日还有哪些重要变化Google DeepMind 推出 Gemini Omni 1.2 Flash，支持生成更长视频

这件事到目前为止

360 says its Tulongfeng security system achieved a 99.4% vulnerability reproduction rate on the CyberGym benchmark, the highest score on the leaderboard. QbitAI reports that three independent evaluations averaged 99.38%. The system combines a security-focused model trained on security data and expert experience with specialized agents that analyze code, use tools, generate proof-of-concept exploits, and verify vulnerabilities. QbitAI also describes a rule Tulongfeng added after a sandbox blocked an attempt to use a test address.