腾讯朱雀实验室推出Deep Puzzling，利用AI技术进行代码防护

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"AI技术不断演进，黑客利用AI来进行网络攻击的事件屡见不鲜，传统攻防手法往往乏力应对，在此背景下，通过AI进行代码防护，开始成为行业的技术趋势。"}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"代码防护技术Deep Puzzling"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"近日，全球顶级的信息安全峰会HITB+Cyberweek 2021正式举办，腾讯朱雀实验室专家研究员Jifeng Zhu和研究员Keyun Luo受邀参加，并进行了题为《Deep Puzzling: Binary Code Intention Hiding based on AI Uninterpretability》（《基于AI不可解释性的二进制代码意图隐藏》）的议题分享。会上，腾讯朱雀实验室展示了如何利用AI模型的特性，实现二进制代码的意图隐藏，有效防止代码被黑客逆向分析，从而保障核心代码的安全。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"据了解，相比传统攻防技术，AI算法具有诸多优势，例如，在复杂特征建模、内容生成、概率容错、不可解释性等方面拥有强大的能力。此次腾讯朱雀实验室推出的Deep Puzzling（深度迷惑）技术正是利用了AI的这些特点，对代码进行深层次的安全布防。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"据介绍，Deep Puzzling通过将多种载荷编码到AI模型的参数中，实现高强度的代码意图隐藏，由此来“迷惑”黑客，令其无法反向分析其中的代码逻辑。这样即使黑客取得了AI模型文件，也很难猜透代码的真实意图。这项技术有效地提高了代码的破解难度，可以帮助更多代码拥有者守护自己的知识产权和信息安全，抑制AI型网络攻击的滋长。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/cc\/ef\/cca434991055af4d5fb98bcdb5f22aef.png","alt":null,"title":"Deep Puzzling核心能力","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"目前，朱雀实验室已将这项技术面向全球开发者开源，方便研究团队灵活取用，用前沿的AI技术助力网络安全的升级。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"},{"type":"strong"}],"text":"Deep Puzzling开源地址："},{"type":"link","attrs":{"href":"https:\/\/github.com\/aisecstudent\/DeepPuzzling","title":null,"type":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"https:\/\/github.com\/aisecstudent\/DeepPuzzling"}],"marks":[{"type":"italic"},{"type":"strong"}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"Deep Puzzling工作原理"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在Deep Puzzling技术框架里，有多个互相连接的模型，通过技术适应，保证让黑客无法通过修改输入数据的方式，来推测输出代码之间的逻辑关联，进而加大了逆向分析代码的难度，提高了核心代码的安全级别。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"不仅如此，朱雀实验室还设计了一个检测模块，来验证这种方法的有效性。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/d4\/c0\/d4f242e76216f0e168c68ce14b9e58c0.png","alt":null,"title":"Deep Puzzling工作原理演示","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，通过读取海量的普通环境数据，构建一个“触发-生成-纠错”模型，端到端地实现了“目标定位-代码执行”步骤，然后直接生成载荷。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"值得一提的是，这个系统还具备反调试能力。这种反调试能力并非传统的进程状态查看、时间分析、异常处理等，而是利用网络构造出没有任何“显式if”判断含义的计算过程，这个计算过程处于黑盒中，很难得知其因果关联性，因而具有良好的数据密封性。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外，由于AI模型产生的代码有一定的错误率，研究员们还设计了一种纠错模型，来进一步降低局部解码的错误率，使得AI模型大概率地输出精确的结果，以确保被计算机正确地执行。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"大量反复稳定性测试有力地佐证了Deep Puzzling的可行性。朱雀实验室的研究员透露，“我们邀请过业界多位资深的逆向工程研究人员来尝试破解，均无法解出，更加验证了这是一个非常值得关注的新方向。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"据了解，早在2018年，就有前人尝试过利用AI技术来完成代码的意图隐藏。当时有研究人员提出了一种基于AI密钥的“包装”思路——DeepLocker，其工作原理为，只有特定目标经过AI模型产生的密钥才能解锁意图代码。这项研究展示了AI在意图隐藏方面的巨大潜力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/75\/35\/75b13b9f3f117a45f0c0f4c32c59e035.png","alt":null,"title":"DeepLocker工作原理","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"不过，由于密钥解密代码的逻辑是暴露的，黑客仍然可以找到过程中的漏洞来盗取核心代码。而随着AI技术与网络安全结合得越来越紧密，基于AI技术的网络攻防手段也在日益完善成熟。"}]}]}